Concrete grouting repair body ultrasonic wave intensity prediction method and system

Abstract

The application discloses a concrete grouting prosthesis ultrasonic wave intensity prediction method and system, relates to the technical field of concrete structure defect grouting repair, ultrasonic wave nondestructive testing and mechanical property prediction, and the method is characterized in that, for the basic compressive strength prediction value output by the obtained basic prediction model, the correction coefficients of various influence factors are determined through single-factor paired calibration test, interaction test is conducted on the factors changing simultaneously, a comprehensive correction coefficient is generated, and the final compressive strength prediction value is calculated; the comprehensive correction coefficient is determined through single-factor paired calibration and interaction deviation test, and over-correction or insufficient correction caused by direct multiplication of multiple correction coefficients can be reduced.

Description

Technical Field

[0001] This invention relates to the fields of grouting repair of defects in concrete structures, ultrasonic non-destructive testing, and mechanical property prediction, specifically to a method and system for predicting the ultrasonic strength of grouting repaired concrete structures. Background Technology

[0002] Defects such as voids, cracks, interface debonding, and back voids may occur in tunnel linings, underground structures, road base courses, and existing concrete components during construction and service. After grouting repair using cement-based grout, cement-water glass composite grout, polyurethane grout, or other composite grouts, it is necessary to determine not only whether the defects have been filled, but also the compressive strength and mechanical recovery level of the repaired structure. Traditional compressive strength tests and core sampling can damage the specimen or structure, and it is difficult to repeat the test at the same location.

[0003] Ultrasonic wave velocity is correlated with the internal density, elastic characteristics, and defect development of concrete, and can therefore be used for indirect prediction of compressive strength. However, existing methods typically employ single-path wave velocity and a single regression equation. Grouted repair structures simultaneously contain the original concrete, grouting material, and interface transition zones; different paths may traverse grouting zones with varying volume fractions. Probe coupling, path location, and local porosity can lead to wave velocity dispersion. Different grout densities, elastic moduli, and setting characteristics, as well as constant-pressure or pulsed grouting processes, will alter the wave velocity response at the same strength. Directly adopting a uniform wave velocity-strength relationship can easily result in unclear model applicability and prediction bias.

[0004] Furthermore, some studies, when plotting the relationship between wave velocity and compressive strength, use compressive strength as the x-axis and wave velocity as the y-axis for regression. However, engineering applications require predicting compressive strength from measured wave velocity. The model parameters for the two regression directions are not equivalent, and simply performing an inverse algebraic solution to the original equation is insufficient to fully reflect the error structure when compressive strength is the prediction target. Therefore, it is necessary to establish a calibration model with compressive strength as the dependent variable, and combine multi-path data screening, benchmark normalization, material and process correction, and prediction reliability assessment. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for predicting the ultrasonic strength of grouting repair bodies in concrete, so as to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for predicting the ultrasonic strength of grouting repair bodies, comprising the following steps: Construct a set of calibration specimens, obtain the measured compressive strength of each calibration specimen, and record the grouting material, aggregate particle size, defect geometric parameters, grouting process, curing age, moisture content and test temperature of each calibration specimen; No fewer than three ultrasonic propagation paths shall be set up in the grouting repair area to be tested. At the same time, reference propagation paths shall be set up on adjacent complete areas or complete reference specimens of the same batch. The ultrasonic propagation time of each path shall be collected repeatedly no fewer than three times. The path representative wave velocity is calculated from the repeated data collected from each path. After removing outliers from the collected values, weights are assigned according to the repeatability index of each path to obtain the effective wave velocity. The normalized wave velocity is then calculated based on the benchmark wave velocity corresponding to the same conditions. Based on the consistency between the calibration system and the test object in terms of batch, age, water content and temperature, one of the obtained effective wave velocity and the corresponding reference wave velocity under the same conditions is selected as the input variable of the basic prediction model; the basic prediction model is selected from multiple candidate models by using a hierarchical cross-validation method, and the predicted value of the basic compressive strength is calculated using the model. For the basic compressive strength prediction value output by the obtained basic prediction model, the correction coefficient of each influencing factor is determined by single-factor paired calibration test, the interaction test of factors that change at the same time is carried out, a comprehensive correction coefficient is generated, and the final compressive strength prediction value is calculated. Based on the coefficient of variation between the wave velocities represented by each propagation path, whether the selected input variables are within the model calibration range, and the prediction error of the model during the verification process, the reliability of the final prediction results is graded and output.

[0007] Furthermore, the defect geometric parameters of the calibration specimen are determined based on the design dimensions and location records of the reserved mold or the embedded defect body; The geometric parameters of defects in existing concrete structures are obtained through at least one of the following methods: construction records, endoscopic inspection, computed tomography, or ultrasonic tomography. The defect geometric parameters include defect type, equivalent size, defect volume ratio, and the proportion of the length of the ultrasonic detection path in the grouting repair area to the total path length. The calibration specimens cover at least three compressive strength grades, with no fewer than three parallel specimens set under each grade. When the number of valid samples is no less than 50 groups, 70% to 80% are stratified according to strength grade and material process combination as the model establishment group and 20% to 30% as the independent verification group. The verification group contains at least one set of specimens in each strength grade and main factor combination. When the number of valid samples is 20 to 49 groups, a stratified five-fold cross-validation method is adopted, with each set of samples being rotated sequentially as one verification data.

[0008] Furthermore, the ultrasonic wave velocity collected in the kth acquisition along the i-th propagation path is denoted as... The Where Li is the length of the i-th propagation path, Let Vi be the propagation time corresponding to the k-th ultrasonic wave velocity collected for the i-th propagation path. After removing abnormal collected values, the arithmetic mean of the effective wave velocities for each path is taken as the representative wave velocity for that path. The representative wave velocity for the i-th propagation path is denoted as Vi.

[0009] Furthermore, abnormal acquisition values ​​are eliminated using the median deviation method, interquartile range method, or relative deviation threshold method. When the relative deviation between the wave velocity of a certain acquisition and the median wave velocity of the same path is greater than a preset value, the acquisition value of that acquisition is eliminated and re-acquisition is performed until no less than 3 valid acquisition values ​​are retained for each path. The effective wave velocity is denoted as Ve, where the value of Ve is equal to the sum of the products of the repeatability index weight assigned to each path and the wave velocity represented by the corresponding path; and the sum of the repeatability index weight assigned to each path is equal to 1. Let λi be the weight assigned to the repetition index of the i-th path. , where CVi is the coefficient of variation of the repeated collection values ​​of the i-th path, and ε is a preset positive constant used to avoid the case where the denominator is zero when CVi is zero in the weight calculation; The reference wave velocity is taken from a complete area or complete reference specimen that has the same concrete mix ratio, the same curing age, the same moisture content, and the same testing temperature as the grouting repair body to be tested. When the object to be tested and the calibration object belong to different batches or are in different states but the corresponding reference wave velocity can be obtained, the corresponding normalized wave velocity is preferentially used as the model input. The corresponding normalized wave velocity is equal to the quotient between the corresponding effective wave velocity and the reference wave velocity under the same conditions. When the test object and the calibration object belong to the same batch and the same state, and the coefficient of variation between the wave velocities of multiple propagation paths in the reference area or between reference specimens in the same batch is less than or equal to the preset threshold, the corresponding effective wave velocity is used as the model input.

[0010] Furthermore, the candidate models include linear models, quadratic polynomial models, exponential models, and power function models; when the effective wave velocity is used as the model input, the linear model is expressed as fc0=a+bVe, where fc0 is the predicted value of the basic compressive strength; Ve is the corresponding effective wave velocity; and a and b are different fitting coefficients. When the normalized wave velocity is used as input, the linear model is expressed as fc0=a′+b′Rv, where Rv is the corresponding normalized wave velocity; a′ and b′ are different fitting coefficients; and Ve and Rv do not appear as independent variables at the same time in the same linear model. Each candidate model is trained and validated using the same data partitioning method. The root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²) of the validation set are calculated respectively. The minimum RMSE is the primary criterion for selection. If the minimum RMSE is less than the preset value relative to the second-lowest order model, the lower-order model with fewer parameters is selected.

[0011] Furthermore, the influencing factors include the type of grouting material, aggregate particle size or defect geometry, grouting process type, curing age, moisture content, and testing temperature; wherein, the grouting material includes cement-based grout, cement-water glass composite grout, polyurethane grout, or composite grout containing expansion components, and the grouting process includes constant pressure grouting, graded pressure grouting, or pulse grouting; When determining the correction coefficients for influencing factors, the concrete batch, specimen size, aggregate particle size, defect type and size, testing path layout, curing age, moisture content, and testing temperature are all kept the same; only the influencing factors themselves are changed. At least three parallel specimens are set up for each factor level, and the correction coefficient for the j-th influencing factor is denoted as Kj. Where fc,meas,k is the measured compressive strength of the kth parallel specimen when only the jth influencing factor itself is changed, fc0,k is the basic predicted value of the kth parallel specimen when only the jth influencing factor itself is changed, and n is the total number of parallel specimens; When the j-th influencing factor and the k-th influencing factor change simultaneously, a combined specimen is prepared separately, and the joint correction coefficient is calculated based on the measured data of the obtained combined specimen, denoted as Kjk; Calculate the interaction deviation between the j-th influencing factor and the k-th influencing factor, denoted as δjk;

[0012]

[0013] Wherein, Kjk is obtained directly from the combined specimen and represents the actual response level under the combined state. The deviation calculated with Kjk as the denominator can determine whether the product of individual coefficients deviates from the actual response; Kk is the correction coefficient of the kth influencing factor; when δjk is less than or equal to 5%, KjKk is used as the combined correction term of the jth and kth influencing factors; when δjk is greater than 5%, Kjk replaces KjKk; the comprehensive correction coefficient is composed of the combination of each independent single-factor correction coefficient and the joint correction coefficient with interaction, and the same influencing factor is only counted once in the comprehensive correction coefficient. The comprehensive correction coefficient is denoted as Kc, and the final predicted compressive strength value is equal to the product of Kc and the obtained predicted basic compressive strength value.

[0014] Furthermore, the wave velocity variation coefficient between paths is calculated as the ratio of the sample standard deviation of the wave velocity representative of each path to its arithmetic mean. When the wave velocity variation coefficient between paths is less than or equal to 3%, the model input is within the calibration range, and the model validation MAPE is less than or equal to 12%, the reliability level is determined to be A. When the inter-path wave velocity variation coefficient is greater than 3% and less than or equal to 5%, the model input is within the calibration range, and the model validation MAPE is less than or equal to 15%, the reliability level is determined to be B. When the wave velocity variation coefficient between paths is greater than 5%, or the model input exceeds the calibration range, or the model verification MAPE is greater than 15%, the reliability level is determined to be C, and prompts are given to recouple the detection surface, add detection paths, or supplement the calibration test.

[0015] Furthermore, the prediction results output shall include at least the following information: original wave velocity and outlier processing records for each path, effective wave velocity, normalized wave velocity, finally selected input variable type, basic model expression and its regression coefficients, correction coefficients for each single factor or joint correction coefficients, final predicted compressive strength value, model calibration scope, current model version number, and reliability level determination result.

[0016] An ultrasonic strength prediction system for grouting repairs includes: The ultrasonic acquisition module is used to collect the length data of each propagation path and the propagation time of ultrasonic waves in each repeated test; The wave velocity calculation and quality control module is used to calculate the representative wave velocity of each path based on the collected data, remove abnormal collected values, and calculate the effective wave velocity after assigning weights according to the repeatability index of each path. The reference normalization and input selection module is used to calculate the normalized wave velocity when a reference wave velocity is available, and to determine the model input variables from the effective wave velocity and the normalized wave velocity based on the consistency of conditions between the calibration object and the object under test. The model selection and prediction module, based on multiple stored candidate models, combines the input calibration data to perform model training, cross-validation and model selection, and calls the selected basic prediction model to calculate the predicted value of basic compressive strength. The correction coefficient determination and cross-validation module is used to determine the correction coefficients for each single factor based on paired calibration test data, perform cross-variance calculations for cases where multiple factors change simultaneously, and generate a comprehensive correction coefficient. The reliability grading and output module is used to grade the reliability of the prediction results based on the wave velocity variation coefficient between paths, whether the input variables are within the calibration range, and the model validation error, and output the grading results and the predicted values ​​together.

[0017] Furthermore, the system is configured with an updatable model library and a correction coefficient library; the model version update and evaluation process is triggered when any of the following occurs: The number of newly added valid calibration samples reaches more than 10 groups or more than 20% of the total number of samples in the original model; The grouting material type, grouting process, or defect geometry combination not covered in the original model calibration data appears; The MAPE of 20 consecutive sets of retest samples in chronological order of detection is greater than 15%, or the average absolute value of the signed error of these 20 sets of samples is greater than 5%. The input variable value of the test object exceeds the upper or lower limit of the original model calibration interval by more than 5%; The version update evaluation process includes the following steps in sequence: reviewing the path repeatability, state record integrity, and reliability of the stress test data for the new data; freezing the independent validation set or fixed cross-validation partitioning method used when the original model was built; incorporating the approved new data into the training set and refitting each candidate model and correction coefficient according to the same rules as the original modeling; comparing the error indices of the new and old models on the same validation dataset; only when the root mean square error (RMSE) of the new model's validation set is reduced by at least 5% compared to the old model, the mean absolute error (MAE) is not higher than that of the old model, and the increase in the mean absolute percentage error (MAPE) of each major factor subgroup does not exceed 10%, will the new version be activated; after the version update, the old version model, its update date, sample range, regression coefficients, and validation indices will be archived and saved to form a traceable version record.

[0018] Compared with the prior art, the beneficial effects achieved by the present invention are: (1) The present invention employs multiple propagation paths and repeated detection, which can reduce the impact of single measuring point, local pores and probe coupling error on wave velocity measurement; (2) The present invention forms an effective wave velocity by removing outliers and using a weighting method based on repetition, so that the data input to the model has a clear quality control process; (3) The present invention uses a complete region or complete reference specimen under the same conditions to normalize the wave velocity, which can reduce the batch differences in concrete mix proportion, age, moisture content and test temperature. (4) The present invention selects either Ve or Rv as the model input based on the baseline conditions to avoid collinearity caused by both entering the linear model at the same time, and directly establishes the prediction model with compressive strength as the dependent variable. (5) The present invention uses a unified data partitioning and verification index to screen candidate models, and prioritizes the selection of low-order models when the performance difference is small, so that the model selection process is clear and reproducible; (6) The present invention determines the comprehensive correction coefficient by single-factor pairing calibration and interaction bias test, which can reduce the overcorrection or undercorrection caused by direct multiplication of multiple correction coefficients; (7) This invention combines path discreteness, model calibration range and verification error to make A, B and C level judgments, and specifies the model update triggering conditions, which can avoid unbounded extrapolation and ensure the traceability of the prediction process; (8) The present invention can repeatedly test the same repair area without damaging the structure, and is suitable for grouting repair scheme comparison, quality acceptance and subsequent status tracking. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the ultrasonic strength prediction method for concrete grouting repair bodies according to the present invention. Figure 2 This is a schematic diagram of the multi-path ultrasonic detection and reference path arrangement in this invention; Figure 3 This is a flowchart of the hierarchical correction process for basic predicted values ​​in this invention; Figure 4 This is a schematic diagram of the method for verifying the predicted compressive strength and the measured compressive strength in this invention; Figure 5 This is a schematic diagram of the calibration specimen parameter recording items in this invention; Figure 6 This is a schematic diagram of the hierarchical five-fold cross-validation results of the candidate model in this invention; Figure 7 This is a schematic diagram illustrating the method for determining the layered correction coefficient in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 This embodiment provides a method for predicting the ultrasonic strength of grouting repairs, including the following steps: Construct a set of calibration specimens, obtain the measured compressive strength of each calibration specimen, and record the grouting material, aggregate particle size, defect geometric parameters, grouting process, curing age, moisture content and test temperature of each calibration specimen; The defect geometric parameters of the calibration specimen are determined based on the design dimensions and location records of the reserved mold or embedded defect body; The geometric parameters of defects in existing concrete structures are obtained through at least one of the following methods: construction records, endoscopic inspection, computed tomography, or ultrasonic tomography. The defect geometric parameters include defect type, equivalent size, defect volume ratio, and the proportion of the length of the ultrasonic detection path in the grouting repair area to the total path length. The calibration specimens cover at least three compressive strength grades, with no fewer than three parallel specimens set under each grade. When the number of valid samples is no less than 50 groups, 70% to 80% are stratified according to strength grade and material process combination as the model establishment group and 20% to 30% as the independent verification group. The verification group contains at least one set of specimens in each strength grade and main factor combination. When the number of valid samples is 20 to 49 groups, a stratified five-fold cross-validation method is adopted, with each set of samples being rotated sequentially as one verification data.

[0022] At least three ultrasonic propagation paths should be laid out in the grouting repair area to be tested, and reference propagation paths should be laid out on adjacent intact areas or intact reference specimens of the same batch. The ultrasonic propagation time of each path should be repeatedly collected at least three times; Figure 2 As shown, transmitting transducer groups and receiving transducer groups are arranged on opposite sides of the grouting repair area to be tested. Four detection paths P1 to P4 are set up that pass through or are adjacent to the grouting repair area, and a reference path is set up in the adjacent intact area. All paths use the same frequency, sampling rate, trigger threshold and coupling agent.

[0023] The path representative wave velocity is calculated from the repeated data collected from each path. After removing outliers from the collected values, weights are assigned according to the repeatability index of each path to obtain the effective wave velocity. The normalized wave velocity is then calculated based on the benchmark wave velocity corresponding to the same conditions. Let the ultrasonic wave velocity collected in the kth acquisition along the i-th propagation path be denoted as... The Where Li is the length of the i-th propagation path, Let Vi be the propagation time corresponding to the k-th ultrasonic wave velocity collected for the i-th propagation path. After removing abnormal collected values, the arithmetic mean of the effective wave velocities for each path is taken as the representative wave velocity for that path. The representative wave velocity for the i-th propagation path is denoted as Vi.

[0024] Abnormal acquisition values ​​are removed using the median deviation method, interquartile range method, or relative deviation threshold method. When the relative deviation between the wave velocity of a certain acquisition and the median wave velocity of the same path is greater than a preset value (5% of the median wave velocity of the same path), the acquisition value is removed and reacquired until at least 3 valid acquisition values ​​are retained for each path. The effective wave velocity is denoted as Ve, where the value of Ve is equal to the sum of the products of the repeatability index weight assigned to each path and the wave velocity represented by the corresponding path; and the sum of the repeatability index weight assigned to each path is equal to 1. Let λi be the weight assigned to the repetition index of the i-th path. , where CVi is the coefficient of variation of the repeated collection values ​​of the i-th path, and ε is a preset positive constant used to avoid the case where the denominator is zero when CVi is zero in the weight calculation; The reference wave velocity is taken from a complete area or complete reference specimen that has the same concrete mix ratio, the same curing age, the same moisture content and the same test temperature as the grouting repair body to be tested. Based on the consistency between the calibration system and the test object in terms of batch, age, water content and temperature, one of the obtained effective wave velocity and the corresponding reference wave velocity under the same conditions is selected as the input variable of the basic prediction model; the basic prediction model is selected from multiple candidate models by using a hierarchical cross-validation method, and the predicted value of the basic compressive strength is calculated using the model. When the test object and the calibration object belong to different batches or are in different states but the corresponding reference wave velocity can be obtained, the corresponding normalized wave velocity is used as the model input first. The corresponding normalized wave velocity is equal to the quotient between the corresponding effective wave velocity and the reference wave velocity under the same conditions. When the test object and the calibration object belong to the same batch and the same state, and the coefficient of variation between the wave velocities of multiple propagation paths in the reference area or between reference specimens in the same batch is less than or equal to the preset threshold (take 3%), the corresponding effective wave velocity is used as the model input.

[0025] The candidate models include linear models, quadratic polynomial models, exponential models, and power function models. When the effective wave velocity is used as the model input, the linear model is expressed as fc0=a+bVe, where fc0 is the predicted value of the basic compressive strength, Ve is the corresponding effective wave velocity, and a and b are different fitting coefficients. When the normalized wave velocity is used as input, the linear model is expressed as fc0=a′+b′Rv, where Rv is the corresponding normalized wave velocity; a′ and b′ are different fitting coefficients; and Ve and Rv do not appear as independent variables at the same time in the same linear model. Each candidate model is trained and validated using the same data partitioning method. The root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²) of the validation set are calculated respectively. The minimum RMSE is the primary criterion for selection. If the minimum RMSE is less than the preset value (5%) of the RMSE reduction of the second-lowest-order model, the lower-order model with fewer parameters is selected.

[0026] For the basic compressive strength prediction value output by the obtained basic prediction model, the correction coefficient of each influencing factor is determined by single-factor paired calibration test, the interaction test of factors that change at the same time is carried out, a comprehensive correction coefficient is generated, and the final compressive strength prediction value is calculated. The influencing factors include the type of grouting material, aggregate particle size or defect geometry, grouting process type, curing age, moisture content, and testing temperature; among which, the grouting material includes cement-based grout, cement-water glass composite grout, polyurethane grout, or composite grout containing expansion components, and the grouting process includes constant pressure grouting, graded pressure grouting, or pulse grouting; When determining the correction coefficients for influencing factors, the concrete batch, specimen size, aggregate particle size, defect type and size, testing path layout, curing age, moisture content, and testing temperature are all kept the same; only the influencing factors themselves are changed. At least three parallel specimens are set up for each factor level, and the correction coefficient for the j-th influencing factor is denoted as Kj. Where fc,meas,k is the measured compressive strength of the kth parallel specimen when only the jth influencing factor itself is changed, fc0,k is the basic predicted value of the kth parallel specimen when only the jth influencing factor itself is changed, and n is the total number of parallel specimens; When the j-th influencing factor and the k-th influencing factor change simultaneously, a combined specimen is prepared separately, and the joint correction coefficient is calculated based on the measured data of the obtained combined specimen, denoted as Kjk; Calculate the interaction deviation between the j-th influencing factor and the k-th influencing factor, denoted as δjk;

[0027]

[0028] Wherein, Kjk is obtained directly from the combined specimen and represents the actual response level under the combined state. Using Kjk as the denominator to calculate the deviation can determine whether the product of individual coefficients deviates from the actual response; Kk is the correction coefficient for the k-th influencing factor; when δjk is less than or equal to 5%, KjKk is used as the combined correction term for the j-th and k-th influencing factors; when δjk is greater than 5%, Kjk replaces KjKk; the comprehensive correction coefficient is composed of the independent single-factor correction coefficients and the joint correction coefficients with interactive effects, and the same influencing factor is only counted once in the comprehensive correction coefficient, denoted as Kc. The final predicted compressive strength value is equal to the product of Kc and the obtained predicted basic compressive strength value, such as... Figure 3 As shown.

[0029] Based on the coefficient of variation between the wave velocities represented by each propagation path, whether the selected input variables are within the model calibration range, and the prediction error during the model validation process, the reliability of the final prediction results is graded and output, such as... Figure 4As shown.

[0030] The wave velocity variation coefficient between paths is calculated as the ratio of the sample standard deviation of the wave velocity representative of each path to its arithmetic mean. When the wave velocity variation coefficient between paths is less than or equal to 3%, the model input is within the calibration range, and the model validation MAPE is less than or equal to 12%, the reliability level is determined to be A. When the inter-path wave velocity variation coefficient is greater than 3% and less than or equal to 5%, the model input is within the calibration range, and the model validation MAPE is less than or equal to 15%, the reliability level is determined to be B. When the wave velocity variation coefficient between paths is greater than 5%, or the model input exceeds the calibration range, or the model verification MAPE is greater than 15%, the reliability level is determined to be C, and prompts are given to recouple the detection surface, add detection paths, or supplement the calibration test.

[0031] The prediction results output shall include at least the following information: original wave velocity and outlier processing records for each path, effective wave velocity, normalized wave velocity, final selected input variable type, basic model expression and its regression coefficients, correction coefficients for each single factor or joint correction coefficients, final predicted compressive strength value, model calibration scope, current model version number, and reliability level determination result.

[0032] This embodiment also provides an ultrasonic strength prediction system for grouting repair bodies, including: The ultrasonic acquisition module is used to collect the length data of each propagation path and the propagation time of ultrasonic waves in each repeated test; The wave velocity calculation and quality control module is used to calculate the representative wave velocity of each path based on the collected data, remove abnormal collected values, and calculate the effective wave velocity after assigning weights according to the repeatability index of each path. The reference normalization and input selection module is used to calculate the normalized wave velocity when a reference wave velocity is available, and to determine the model input variables from the effective wave velocity and the normalized wave velocity based on the consistency of conditions between the calibration object and the object under test. The model selection and prediction module, based on multiple stored candidate models, combines the input calibration data to perform model training, cross-validation and model selection, and calls the selected basic prediction model to calculate the predicted value of basic compressive strength. The correction coefficient determination and cross-validation module is used to determine the correction coefficients for each single factor based on paired calibration test data, perform cross-variance calculations for cases where multiple factors change simultaneously, and generate a comprehensive correction coefficient. The reliability grading and output module is used to grade the reliability of the prediction results based on the wave velocity variation coefficient between paths, whether the input variables are within the calibration range, and the model validation error, and output the grading results and the predicted values ​​together.

[0033] The system is equipped with an updatable model library and correction coefficient library; the model version update and evaluation process is triggered when any of the following occurs: The number of newly added valid calibration samples reaches more than 10 groups or more than 20% of the total number of samples in the original model; The grouting material type, grouting process, or defect geometry combination not covered in the original model calibration data appears; The MAPE of 20 consecutive sets of retest samples in chronological order of detection is greater than 15%, or the average absolute value of the signed error of these 20 sets of samples is greater than 5%. The input variable value of the test object exceeds the upper or lower limit of the original model calibration interval by more than 5%; The version update evaluation process includes the following steps in sequence: reviewing the path repeatability, state record integrity, and reliability of the stress test data for the new data; freezing the independent validation set or fixed cross-validation partitioning method used when the original model was built; incorporating the approved new data into the training set and refitting each candidate model and correction coefficient according to the same rules as the original modeling; comparing the error indices of the new and old models on the same validation dataset; only when the root mean square error (RMSE) of the new model's validation set is reduced by at least 5% compared to the old model, the mean absolute error (MAE) is not higher than that of the old model, and the increase in the mean absolute percentage error (MAPE) of each major factor subgroup does not exceed 10%, will the new version be activated; after the version update, the old version model, its update date, sample range, regression coefficients, and validation indices will be archived and saved to form a traceable version record. Example 1:

[0034] 100 mm × 100 mm × 100 mm concrete cube specimens were prepared, and spherical voids, cracks, or interface debonding defects were created inside some specimens. These defects were then repaired using cement-based grout, cement-water glass composite grout, or polyurethane grout. Cement-based grout can be prepared with a cement-to-water ratio of 1:0.6, while cement-water glass composite grout can be prepared with a cement, water, and water glass ratio of 1:0.6:1.6. Grouting techniques included constant pressure grouting and pulse grouting.

[0035] The calibration specimens should cover at least three compressive strength grades, with at least three parallel specimens for each grade. For spherical voids, record the equivalent diameter and volume fraction; for cracks, record the width, length, and orientation; for interface debonding, record the debonding area ratio. The path crossing rate is obtained by dividing the length of the path within the grouting repair zone by the total path length. After reaching the predetermined age, ultrasonic testing is performed first, followed by compressive strength testing on the corresponding specimens to generate a one-to-one correspondence between wave velocity and compressive strength calibration data. When the sample size is no less than 50 groups, stratified model establishment and independent verification groups are used; for 20–49 groups, stratified five-fold cross-validation is employed. Key recorded items include… Figure 5 As shown.

[0036] Transmitter and receiver groups are arranged on opposite sides of the grouting repair area to be tested. Four detection paths P1 to P4 are set up that pass through or are adjacent to the grouting repair area, and a reference path is set up in the adjacent intact area. All paths use the same frequency, sampling rate, trigger threshold and coupling agent.

[0037] To illustrate the calculation process, assume that the average wave velocities of the five effective paths for a certain specimen are 3688 m / s, 3701 m / s, 3712 m / s, 3695 m / s, and 3704 m / s, respectively, and that the coefficient of variation for each path in repeated detections does not exceed 1%. After calculation using the weighting formula, the effective wave velocity is approximately 3700 m / s. If the effective wave velocity of a complete reference specimen under the same conditions is 4210 m / s, then the normalized wave velocity Rv is approximately 0.879.

[0038] Regression analysis was performed on the wave velocity and measured compressive strength of the calibrated specimens. First, either Ve or Rv was selected as the input variable based on the availability of complete reference standards and batch differences. Then, the same data partitioning was used to compare linear, quadratic polynomial, exponential, and power function models. This example has 32 sets of data, less than 50 sets, therefore stratified five-fold cross-validation was performed according to strength range and data source, with each set serving as a validation sample. The model prioritizes the validation RMSE, while also comparing MAE and MAPE; if the lowest RMSE decreases by less than 5% relative to the next lower-order model, the lower-order model with fewer parameters is selected, such as... Figure 6 As shown, the exponential model only reduced the validation RMSE by about 2.9% compared to the linear model, which did not reach the 5% complexity gain threshold. Therefore, the linear model was chosen as the basic prediction model in this embodiment, and the final coefficients were refitted using all 32 sets of calibration data.

[0039] This embodiment uses a total of 32 sets of wave velocity-compressive strength correspondence data, including 15 sets of data from grouting repair specimens tested in this experiment and 17 sets of pre-prepared comparative data from publicly available literature. Using ultrasonic wave velocity as the independent variable and compressive strength as the dependent variable, a linear prediction model is re-established using the least squares method.

[0040]

[0041] The model has a coefficient of determination (R²) of 0.75736, a root mean square error (RMSE) of 5.25 MPa, and a mean absolute error (MAE) of 3.97 MPa. The model is calibrated for wave velocities ranging from 2612.80 to 5058.51 m / s, corresponding to compressive strength ranges of 20.72 to 58.54 MPa. In the formula, Ve is in m / s and fc0 is in MPa. This model is used to illustrate the direct regression modeling process of this invention, using wave velocity as input and compressive strength as output. When applied to other material systems, specimen sizes, or testing conditions, the model coefficients should be re-determined using the corresponding calibration data. When the effective wave velocity Ve is 3700 m / s, the predicted basic compressive strength is fc0 = 0.99527 + 0.00968665 × 3700 ≈ 36.84 MPa. If the object under test is consistent with the basic calibration group in terms of material, geometry, process, age and moisture content, then all correction coefficients are taken as 1, and the final predicted value is still about 36.84 MPa.

[0042] Taking grouting material correction as an example, specimens with identical dimensions, aggregate size, defect type and size, path arrangement, grouting pressure, curing age, moisture content, and testing temperature are prepared from the same concrete batch. Only the grout type is changed, with at least three parallel specimens for each grout type. The ratio of the measured strength to the predicted strength of the foundation is calculated for the k-th specimen, and the material correction coefficient Km is taken as the average of this ratio. The correction coefficients for process, curing age, moisture content, and defect geometry are determined using the same single-factor pairing method, such as... Figure 7 As shown, when two factors change simultaneously, a corresponding combination of specimens is set up and the joint correction coefficient Kjk is calculated. Taking an illustrative example, the material coefficient Km = 1.05 and the process coefficient Kp = 1.03, their product is 1.0815. If the combined specimen yields Kmp = 1.15, then the interaction deviation δmp = |1.15−1.0815| / 1.15×100% = 5.96%, which is greater than 5%. Therefore, Kmp = 1.15 is used instead of KmKp to avoid insufficient correction due to direct multiplication. The correction coefficient and the applicable scope are stored together in the coefficient library; combinations of factors not covered are not extrapolated from adjacent categories, but are supplemented with calibration or have their reliability level reduced.

[0043] The coefficient of variation (CVp) of wave velocity between paths is calculated using the representative wave velocity Vi for each path, i.e., CVp = s(V1, V2, ..., Vn) / V × 100%. Taking the five paths in the above embodiment as an example, the average path speed is 3700 m / s, and the CVp is approximately 0.25%. This input is within the calibration range of 2612.80 to 5058.51 m / s, and the layered five-fold verification MAPE in this embodiment is 10.80%, not exceeding 12%. When all corresponding correction coefficients are complete, it is judged as Grade A. If 3% < CVp ≤ 5% and the verification MAPE is not greater than 15%, it is judged as Grade B; if CVp is greater than 5%, the input exceeds the calibration range, or the verification MAPE is greater than 15%, it is judged as Grade C, and the direct output of the final engineering prediction value is stopped.

[0044] For Grade A and Grade B results, the system synchronously outputs the predicted value, CVp, model input and its calibration range, validation RMSE, MAE and MAPE, the called correction coefficients, and the model version. For Grade C results, if CVp exceeds the limit, the coupling surface is first cleaned again and the probe pressure is kept consistent before repeating the test; if it still exceeds the limit, an additional path is added. When the input exceeds the calibration range, calibration specimens covering the corresponding wave velocity and intensity range are added, and no boundless extrapolation is performed.

[0045] In the 32 sets of data in this embodiment, the experimental data and the pre-defined publicly available data are identified by different symbols and are used together to demonstrate the regression process of directly predicting compressive strength from wave velocity. The publicly available data is used to expand the coverage of strength and wave velocity and to verify the overall trend; in specific engineering applications, the basic model parameters are preferably determined by the data of self-calibrated specimens that are the same as or similar to the test object in terms of material, size, age and testing status.

[0046] Model version updates are triggered by any of the following conditions: At least 10 new valid calibration samples are added, or the sample size reaches 20% of the original model; a grouting material, process, or defect geometric combination not covered by the original model appears; the MAPE of 20 consecutive verification samples is greater than 15% or the mean absolute value of the signed error is greater than 5%; the input to be tested exceeds the original calibration range by more than 5%. After triggering, the following steps are executed sequentially: ① Review the path repeatability, status records, and integrity of the compressive strength test for the new data; ② Freeze the independent validation set or fixed cross-validation partition used by the original model; ③ Merge qualified data and refit candidate models and correction coefficients according to the same rules; ④ Compare the old and new models on the same validation data; ⑤ Activate the new version only when the RMSE of the new model decreases by at least 5%, the MAE does not increase, and the MAPE of any major factor subgroup does not increase by more than 10%; ⑥ Save the old model, update date, sample range, coefficients, and validation indicators to form a traceable version record.

[0047] The data records output by the prediction system should include at least the specimen or test area number, the original wave velocity for each path, the outlier handling results, the effective wave velocity, the reference wave velocity, the normalized wave velocity, the basic predicted value, each correction coefficient, the final predicted value, the reliability level, and the model version number, so as to verify and trace the prediction process.

[0048] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0049] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the ultrasonic strength of grouting repair bodies, characterized in that, Includes the following steps: Construct a set of calibration specimens, obtain the measured compressive strength of each calibration specimen, and record the grouting material, aggregate particle size, defect geometric parameters, grouting process, curing age, moisture content and test temperature of each calibration specimen; No fewer than three ultrasonic propagation paths shall be set up in the grouting repair area to be tested. At the same time, reference propagation paths shall be set up on adjacent complete areas or complete reference specimens of the same batch. The ultrasonic propagation time of each path shall be collected repeatedly no fewer than three times. The path representative wave velocity is calculated from the repeated data collected from each path. After removing outliers from the collected values, weights are assigned according to the repeatability index of each path to obtain the effective wave velocity. The normalized wave velocity is then calculated based on the benchmark wave velocity corresponding to the same conditions. Based on the consistency between the calibration system and the test object in terms of batch, age, water content and temperature, one of the obtained effective wave velocity and the corresponding reference wave velocity under the same conditions is selected as the input variable of the basic prediction model; the basic prediction model is selected from multiple candidate models by using a hierarchical cross-validation method, and the predicted value of the basic compressive strength is calculated using the model. For the basic compressive strength prediction value output by the obtained basic prediction model, the correction coefficient of each influencing factor is determined by single-factor paired calibration test, the interaction test of factors that change at the same time is carried out, a comprehensive correction coefficient is generated, and the final compressive strength prediction value is calculated. Based on the coefficient of variation between the wave velocities represented by each propagation path, whether the selected input variables are within the model calibration range, and the prediction error of the model during the verification process, the reliability of the final prediction results is graded and output.

2. The method for predicting the ultrasonic strength of grouting repair bodies according to claim 1, characterized in that, The defect geometric parameters of the calibration specimen are determined based on the design dimensions and location records of the reserved mold or embedded defect body; The geometric parameters of defects in existing concrete structures are obtained through at least one of the following methods: construction records, endoscopic inspection, computed tomography, or ultrasonic tomography. The defect geometric parameters include defect type, equivalent size, defect volume ratio, and the proportion of the length of the ultrasonic detection path in the grouting repair area to the total path length. The calibration specimens cover at least three compressive strength grades, with no fewer than three parallel specimens set under each grade. When the number of valid samples is no less than 50 groups, 70% to 80% are stratified according to strength grade and material process combination as the model establishment group and 20% to 30% as the independent verification group. The verification group contains at least one set of specimens in each strength grade and main factor combination. When the number of valid samples is 20 to 49 groups, a stratified five-fold cross-validation method is adopted, with each set of samples being rotated sequentially as one verification data.

3. The method for predicting the ultrasonic strength of grouting repair bodies according to claim 1, characterized in that, Let the ultrasonic wave velocity collected in the kth acquisition along the i-th propagation path be denoted as... The Where Li is the length of the i-th propagation path, Let Vi be the propagation time corresponding to the k-th ultrasonic wave velocity collected for the i-th propagation path. After removing abnormal collected values, the arithmetic mean of the effective wave velocities for each path is taken as the representative wave velocity for that path. The representative wave velocity of the i-th propagation path is denoted as Vi.

4. The method for predicting the ultrasonic strength of grouting repair bodies according to claim 1, characterized in that, Abnormal acquisition values ​​are removed using the median deviation method, interquartile range method, or relative deviation threshold method. When the relative deviation between the wave velocity of a certain acquisition and the median wave velocity of the same path is greater than a preset value, the acquisition value of that acquisition is removed and reacquired until no less than 3 valid acquisition values ​​are retained for each path. The effective wave velocity is denoted as Ve, where the value of Ve is equal to the sum of the products of the repeatability index weight assigned to each path and the wave velocity represented by the corresponding path; and the sum of the repeatability index weight assigned to each path is equal to 1. Let λi be the weight assigned to the repetition index of the i-th path. , where CVi is the coefficient of variation of the repeated collection values ​​of the i-th path, and ε is a preset positive constant used to avoid the case where the denominator is zero when CVi is zero in the weight calculation; The reference wave velocity is taken from a complete area or complete reference specimen that has the same concrete mix ratio, the same curing age, the same moisture content, and the same testing temperature as the grouting repair body to be tested. When the object to be tested and the calibration object belong to different batches or are in different states but the corresponding reference wave velocity can be obtained, the corresponding normalized wave velocity is preferentially used as the model input. The corresponding normalized wave velocity is equal to the quotient between the corresponding effective wave velocity and the reference wave velocity under the same conditions. When the test object and the calibration object belong to the same batch and the same state, and the coefficient of variation between the wave velocities of multiple propagation paths in the reference area or between reference specimens in the same batch is less than or equal to the preset threshold, the corresponding effective wave velocity is used as the model input.

5. The method for predicting the ultrasonic strength of grouting repair bodies according to claim 1, characterized in that, The candidate models include linear models, quadratic polynomial models, exponential models, and power function models. When the effective wave velocity is used as the model input, the linear model is expressed as fc0=a+bVe, where fc0 is the predicted value of the basic compressive strength, Ve is the corresponding effective wave velocity, and a and b are different fitting coefficients. When the normalized wave velocity is used as input, the linear model is expressed as fc0=a′+b′Rv, where Rv is the corresponding normalized wave velocity; a′ and b′ are different fitting coefficients; and Ve and Rv do not appear as independent variables at the same time in the same linear model. Each candidate model is trained and validated using the same data partitioning method. The root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²) of the validation set are calculated respectively. The minimum RMSE is the primary criterion for selection. If the minimum RMSE is less than the preset value relative to the second-lowest order model, the lower-order model with fewer parameters is selected.

6. The method for predicting the ultrasonic strength of grouting repair bodies according to claim 1, characterized in that, The influencing factors include the type of grouting material, aggregate particle size or defect geometry, grouting process type, curing age, moisture content, and testing temperature; among which, the grouting material includes cement-based grout, cement-water glass composite grout, polyurethane grout, or composite grout containing expansion components, and the grouting process includes constant pressure grouting, graded pressure grouting, or pulse grouting; When determining the correction coefficients for influencing factors, the concrete batch, specimen size, aggregate particle size, defect type and size, testing path layout, curing age, moisture content, and testing temperature are all kept the same; only the influencing factors themselves are changed. At least three parallel specimens are set up for each factor level, and the correction coefficient for the j-th influencing factor is denoted as Kj. Where fc,meas,k is the measured compressive strength of the kth parallel specimen when only the jth influencing factor itself is changed, fc0,k is the basic predicted value of the kth parallel specimen when only the jth influencing factor itself is changed, and n is the total number of parallel specimens; When the j-th influencing factor and the k-th influencing factor change simultaneously, a combined specimen is prepared separately, and the joint correction coefficient is calculated based on the measured data of the obtained combined specimen, denoted as Kjk; Calculate the interaction deviation between the j-th influencing factor and the k-th influencing factor, denoted as δjk; Wherein, Kk is the correction coefficient of the kth influencing factor; when δjk is less than or equal to 5%, KjKk is used as the combined correction term of the jth and kth influencing factors; when δjk is greater than 5%, Kjk replaces KjKk; the comprehensive correction coefficient is composed of the combination of each independent single-factor correction coefficient and the joint correction coefficient with interaction, and the same influencing factor is only counted once in the comprehensive correction coefficient. The comprehensive correction coefficient is denoted as Kc, and the final predicted compressive strength value is equal to the product of Kc and the obtained predicted basic compressive strength value.

7. The method for predicting the ultrasonic strength of grouting repair bodies according to claim 5, characterized in that, The wave velocity variation coefficient between paths is calculated as the ratio of the sample standard deviation of the wave velocity representative of each path to its arithmetic mean. When the wave velocity variation coefficient between paths is less than or equal to 3%, the model input is within the calibration range, and the model validation MAPE is less than or equal to 12%, the reliability level is determined to be A. When the inter-path wave velocity variation coefficient is greater than 3% and less than or equal to 5%, the model input is within the calibration range, and the model validation MAPE is less than or equal to 15%, the reliability level is determined to be B. When the wave velocity variation coefficient between paths is greater than 5%, or the model input exceeds the calibration range, or the model verification MAPE is greater than 15%, the reliability level is determined to be C, and prompts are given to recouple the detection surface, add detection paths, or supplement the calibration test.

8. The method for predicting the ultrasonic strength of grouting repair bodies according to claim 1, characterized in that, The prediction results output shall include at least the following information: original wave velocity and outlier processing records for each path, effective wave velocity, normalized wave velocity, final selected input variable type, basic model expression and its regression coefficients, correction coefficients for each single factor or joint correction coefficients, final predicted compressive strength value, model calibration scope, current model version number, and reliability level determination result.

9. An ultrasonic strength prediction system for grouting repairs, characterized in that, include: The ultrasonic acquisition module is used to collect the length data of each propagation path and the propagation time of ultrasonic waves in each repeated test; The wave velocity calculation and quality control module is used to calculate the representative wave velocity of each path based on the collected data, remove abnormal collected values, and calculate the effective wave velocity after assigning weights according to the repeatability index of each path. The reference normalization and input selection module is used to calculate the normalized wave velocity when a reference wave velocity is available, and to determine the model input variables from the effective wave velocity and the normalized wave velocity based on the consistency of conditions between the calibration object and the object under test. The model selection and prediction module, based on multiple stored candidate models, combines the input calibration data to perform model training, cross-validation and model selection, and calls the selected basic prediction model to calculate the predicted value of basic compressive strength. The correction coefficient determination and cross-validation module is used to determine the correction coefficients for each single factor based on paired calibration test data, perform cross-variance calculations for cases where multiple factors change simultaneously, and generate a comprehensive correction coefficient. The reliability grading and output module is used to grade the reliability of the prediction results based on the wave velocity variation coefficient between paths, whether the input variables are within the calibration range, and the model validation error, and output the grading results and the predicted values ​​together.

10. The ultrasonic strength prediction system for grouting repair bodies according to claim 9, characterized in that, The system is equipped with an updatable model library and correction coefficient library; the model version update and evaluation process is triggered when any of the following occurs: The number of newly added valid calibration samples reaches more than 10 groups or more than 20% of the total number of samples in the original model; The grouting material type, grouting process, or defect geometry combination not covered in the original model calibration data appears; The MAPE of 20 consecutive sets of retest samples in chronological order of detection is greater than 15%, or the average absolute value of the signed error of these 20 sets of samples is greater than 5%. The input variable value of the test object exceeds the upper or lower limit of the original model calibration interval by more than 5%; The version update evaluation process includes the following steps in sequence: reviewing the path repeatability, state record integrity, and reliability of the stress test data for the new data; freezing the independent validation set or fixed cross-validation partitioning method used when the original model was built; incorporating the approved new data into the training set and refitting each candidate model and correction coefficient according to the same rules as the original modeling; comparing the error indices of the new and old models on the same validation dataset; only when the root mean square error (RMSE) of the new model's validation set is reduced by at least 5% compared to the old model, the mean absolute error (MAE) is not higher than that of the old model, and the increase in the mean absolute percentage error (MAPE) of each major factor subgroup does not exceed 10%, will the new version be activated; after the version update, the old version model, its update date, sample range, regression coefficients, and validation indices will be archived and saved to form a traceable version record.

Images