Method for detecting internal defects of concrete structure

Abstract

The invention relates to the technical field of civil engineering nondestructive testing, and discloses a method for detecting internal defects of a concrete structure. An excitation source and multiple sound pressure sensors are arranged on the surface of concrete, after sound velocity is calibrated in multiple paths in an intact area, a sound pressure sequence of unified sampling is obtained through unit impact excitation, a failure channel is removed, and maximum time delay is extracted from discrete cross correlation through any sensor and converted into an equivalent propagation path difference matrix. The method comprises the following steps: dispersing a volume to be detected into voxels, constructing a propagation contribution function in combination with each voxel and a sensor geometric path, establishing an optimization model with minimum residual error and sparse constraint, carrying out sparse reconstruction and statistical clustering on voxel impedance equivalent path area contribution, automatically identifying and quantifying positions, shapes and sudden change intensity of internal defects, and carrying out quantitative analysis. Errors caused by sound velocity heterogeneity are effectively suppressed, and imaging precision, reliability and automation level are improved.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology in civil engineering, specifically a method for detecting internal defects in concrete structures. Background Technology

[0002] With the continuous expansion of infrastructure construction in my country, concrete structures (such as bridge piers, tunnels, and pile foundations) are widely used in transportation and civil engineering. During long-term service, concrete structures are highly susceptible to internal defects such as cracks, honeycombing, voids, and delamination due to factors such as material aging, environmental erosion, load variations, and construction quality. These internal defects not only affect the structure's load-bearing capacity and durability but can also lead to serious safety accidents. Therefore, early, accurate, and non-destructive detection and location of internal defects in concrete structures has become a crucial technical requirement for ensuring the healthy and safe operation of civil engineering structures.

[0003] Currently, the detection of internal defects in concrete structures mainly relies on non-destructive testing technologies such as ultrasonic testing, impact echo method, X-ray imaging, and radar detection. Among these, ultrasonic testing is the most widely used in engineering practice due to its strong sensitivity to anomalies such as internal material delamination, cracks, and voids. Traditional ultrasonic testing methods generally employ point-to-point sensor deployment, measuring the propagation delay, amplitude attenuation, and spectral characteristics of sound waves in the concrete medium to infer the location and nature of internal defects. Some methods combine two-dimensional or three-dimensional reconstruction, obtaining the approximate distribution of defect areas through X-ray beam intersection and time-of-flight imaging. However, due to the typically large volume and high heterogeneity of concrete structures, and the highly irregular and spatially dispersed nature of internal defects, existing ultrasonic testing solutions face the following technical bottlenecks: First, conventional point-to-point measurements require high sensor array density and sound source localization accuracy, resulting in complex system setup. On-site deployment struggles to balance coverage and detection resolution, easily leading to blind spots and missed or misjudged defects. Secondly, traditional imaging and localization methods mostly rely on empirical parameters or simple interpolation reconstruction, making it difficult to fully utilize the sparsity and global information in the multipath propagation of sound waves. This limits the accuracy of defect localization and weakens the ability to distinguish between multiple interfering defects. Furthermore, environmental noise interference is significant under complex working conditions, and the heterogeneity of concrete causes spatial variations in sound velocity. Conventional methods struggle to achieve accurate sound velocity calibration and dynamic compensation, affecting the reliability and repeatability of detection results. Moreover, many existing technologies fail to robustly handle outliers or extreme signal fluctuations during data processing, making them susceptible to interference from sporadic noise and external shocks, resulting in low detection stability. In recent years, although some scholars have proposed methods based on digital imaging, beamforming, and full-field inversion to improve detection resolution and sensitivity, these techniques often rely on complex inversion models or large-scale computations, resulting in poor algorithm feasibility and limited adaptability to actual engineering conditions.

[0004] Therefore, this paper aims to propose a method for detecting internal defects in concrete structures. First, an excitation source and multiple sound pressure receivers are deployed on the surface of the structure, and the sound velocity parameters are accurately obtained by calibrating the propagation path. Then, an impact excitation is triggered, and a time-domain sound pressure sampling sequence is collected. The propagation delay difference of each sensor pair is estimated based on the discrete cross-correlation method. Next, in a local three-dimensional rectangular coordinate system, the volume to be detected is divided into grid voxels, a propagation contribution function is constructed, and the initial value of the equivalent contribution of voxel impedance is set by combining the overall average path length. Then, a minimum residual optimization objective with sparse constraints is constructed, and sparse reconstruction is performed to obtain the impedance contribution of each voxel. Finally, through statistical analysis and geometric proximity clustering, the defect region is identified and output. Summary of the Invention

[0005] This invention provides a method for detecting internal defects in concrete structures, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a method for detecting internal defects in concrete structures, comprising: Select a detection area on the surface of the concrete structure, set up an excitation source and no less than two sound pressure receiving sensors, set up a calibrated propagation path in the intact area, record the path length and arrival time, calculate the equivalent sound velocity of each sensor based on the length difference and time difference, and take the average to obtain the reference sound velocity of the detection area. The excitation source is triggered to emit a unit impact excitation signal, and each sensor collects sound pressure according to uniform sampling parameters to form a discrete-time sound pressure sampling sequence, and the failed sensors with all-zero sequences are eliminated; For any two sensor sequences, perform discrete cross-correlation within a preset time delay range to obtain the maximum cross-correlation time delay, convert it into equivalent propagation path difference according to the reference sound speed in the detection area, and obtain the equivalent propagation path difference matrix of all sensor pairs. Establish a local three-dimensional rectangular coordinate system, divide the volume to be detected into multiple voxel units, record the geometric center of the voxel unit and the coordinates of each sensor, and construct the propagation contribution function based on the geometric path length; Based on the propagation contribution function and geometric path distribution, and combined with the overall average path length index, an initial value for the impedance equivalent path area contribution is set for each voxel element, and a variable set is established. The overall optimization objective function is constructed with residual minimization and sparsity constraints. Sparse reconstruction is carried out using the equivalent propagation path difference matrix and propagation contribution function to obtain the estimated equivalent path area contribution of each voxel element impedance. The impedance equivalent path area contribution of each voxel element is statistically analyzed, and the mean and average absolute deviation index are calculated. Abnormal voxel elements are identified according to the degree of deviation. Abnormal voxel units are clustered based on geometric proximity to form defect regions, and the output numbers, voxel unit number sets, and average mutation intensity are generated.

[0007] Optionally, the step of selecting a detection area on the concrete structure surface, deploying an excitation source and at least two sound pressure receiving sensors, setting up a calibrated propagation path in an intact area, recording the path length and arrival time, calculating the equivalent sound velocity of each sensor based on the length difference and time difference, and averaging the results to obtain the reference sound velocity of the detection area, specifically includes: A detection area is selected on the surface of the concrete bridge pier, an excitation source is set up, and more than two sound pressure receiving sensors are arranged at equal intervals along a predetermined direction on the surface of the selected detection area. Each sound pressure receiving sensor is numbered, and the total number of numbers is the total number of sound pressure receiving sensors in the detection area. In the known intact area of ​​the bridge pier, two fixed calibration propagation paths are selected for each sound pressure receiving sensor. The actual physical propagation path lengths of the two calibration propagation paths are obtained, and the arrival time of the sound wave along the two calibration propagation paths to the corresponding sound pressure receiving sensor is recorded. For each sound pressure receiving sensor, the arrival time difference corresponding to the two calibration propagation paths is compared. When the arrival times of the two calibration propagation paths are the same or the time difference is zero, the calibration result is determined to be invalid. Two calibration propagation paths that meet the geometric difference are selected again near the corresponding sound pressure receiving sensor or the sound wave arrival time data is collected again until two calibration propagation paths with different arrival times are obtained. After obtaining two calibration propagation paths with different arrival times, the equivalent sound velocity of the corresponding sound pressure receiving sensor is calculated based on the difference in actual physical length of the two calibration propagation paths and the corresponding difference in arrival time. When the calculation result is not greater than zero, the calibration result is determined to be invalid, and a new calibration propagation path is selected or data is collected again until an equivalent sound velocity greater than zero is obtained. After calculating the equivalent sound velocity for all sound pressure receiving sensors, the arithmetic mean of all equivalent sound velocities is taken to obtain the reference sound velocity for the detection area.

[0008] Optionally, the triggering excitation source emits a unit impact excitation signal, and each sensor collects sound pressure according to uniform sampling parameters to form a discrete-time sound pressure sampling sequence. Failed sensors with all-zero sequences are eliminated. Specifically, this includes: At a preset initial moment, the excitation source is triggered to input a unit impact excitation signal into the concrete structure. The excitation signal has a non-zero amplitude only at the initial moment in the time domain, and the amplitude is zero at other moments. Each sound pressure receiving sensor within the detection area continuously acquires sound pressure response signals from the moment of excitation triggering. During the entire acquisition period, the continuous time function of the sound pressure change of the sound pressure receiving sensor with time is recorded, which constitutes the time domain sound pressure response of each sound pressure receiving sensor. For all sound pressure receiving sensors, a unified sampling frequency and total acquisition time are set. Based on the unified sampling frequency, a discrete sampling time sequence with equal intervals is generated within the acquisition time. The continuous-time sound pressure response of each sound pressure receiving sensor is sampled at each discrete sampling time to form a discrete-time sound pressure sampling sequence for each sound pressure receiving sensor, and the total number of discrete sampling points contained in each sequence is determined. The discrete sampling sequences of all sound pressure receiving sensors are checked one by one. When the sound pressure sampling value of a certain sound pressure receiving sensor is zero at all discrete sampling times, the corresponding sound pressure receiving sensor is determined to be a failed sensor. After redeployment or replacement, the excitation and acquisition operation is performed again until all sound pressure receiving sensors have at least one non-zero sound pressure sampling point during the acquisition period.

[0009] Optionally, the step of performing discrete cross-correlation on any two sensor sequences within a preset time delay range to obtain the maximum cross-correlation time delay, converting it into an equivalent propagation path difference according to the reference sound velocity in the detection area, and obtaining the equivalent propagation path difference matrix for all sensor pairs specifically includes: Within the maximum analysis delay range determined by a unified sampling frequency, a discrete delay sequence is constructed for any two sound pressure receiving sensors with different numbers. The discrete delay sequence consists of a set of delay candidate values ​​with the sampling time interval as the step size, and the delay index covers the range from the maximum negative delay to the maximum positive delay. For each pair of sound pressure receiving sensors, a candidate time delay is selected from the discrete time delay sequence. One of the sound pressure sampling sequences is time-shifted relative to the other sound pressure sampling sequence under the selected candidate time delay. The results are multiplied point by point at all multiplicable sampling points and accumulated to obtain the discrete cross-correlation value under the selected candidate time delay. The time shift, point-by-point multiplication and accumulation operations are performed on all candidate time delays of each pair of sound pressure receiving sensors to form a sequence of discrete cross-correlation functions with respect to time delay. For each pair of sound pressure receiving sensors, the maximum value of the cross-correlation value is found from the discrete cross-correlation function sequence. Among all the candidate delays that obtain the maximum cross-correlation value, the candidate delay with the smallest absolute value is selected first. When there are two symmetrical candidate delays, the candidate delay with a time value not less than zero is selected from these two candidate delays. The selected candidate delay is taken as the maximum correlation delay of the corresponding sensor pair. For each pair of sound pressure receiving sensors, the maximum correlation delay is converted into the sound wave propagation path difference using the reference sound velocity in the detection area and the corresponding maximum correlation delay, forming the equivalent propagation path difference data of the corresponding sensor pair; the same conversion process is performed on all remaining sensor pairs to obtain the equivalent propagation path difference matrix of all sensor pairs.

[0010] Optionally, the step of establishing a local three-dimensional Cartesian coordinate system, dividing the volume to be detected into multiple voxel units, recording the geometric center of each voxel unit and the coordinates of each sensor, and constructing a propagation contribution function based on the geometric path length specifically includes: A local three-dimensional rectangular coordinate system is established in the pier inspection area. Any point on the edge of the inspection area is selected as the origin of the coordinate system, and the three coordinate axes are extended along the width direction, height direction and concrete structure thickness direction of the pier, respectively, to form a spatial reference coordinate system. The area to be detected is divided into multiple voxel units along the three coordinate axes according to a preset center spacing. The center spacing in the three directions is the center spacing along the width direction, the center spacing along the height direction, and the center spacing along the thickness direction, respectively. Each voxel unit is assigned a unique number, and the coordinates of the geometric center of each voxel unit in the three-dimensional coordinate system are recorded. The installation position of each sound pressure receiving sensor in the detection area is mapped to a three-dimensional coordinate system, and the spatial coordinates of each sound pressure receiving sensor are recorded. For each voxel unit, based on the three-dimensional distance relationship between the voxel geometric center and the spatial coordinates of each sound pressure receiving sensor, the geometric path length from the voxel geometric center to each sound pressure receiving sensor is calculated. The geometric path length is the spatial straight-line distance in the three-dimensional rectangular coordinate system. For any pair of sound pressure receiving sensors and any voxel unit, the geometric path lengths from the voxel unit to the two sensors are summed. Based on the total path length, a propagation contribution weight is constructed. By setting the reciprocal of the total path length as the weight value, or setting the function value that is monotonically inversely proportional to the total path length as the weight value, the contribution coefficient of the voxel unit to the corresponding sensor to the propagation path is obtained, forming a propagation contribution function about the voxel number and the sensor pair number.

[0011] Optionally, based on the propagation contribution function and geometric path distribution, and combined with the overall average path length index, an initial value for the impedance equivalent path area contribution of each voxel element is set, and a variable set is established. Specifically, this includes: A scalar parameter is set for each voxel unit in the detection area, and the set scalar parameter is defined as the impedance equivalent path area contribution of the voxel unit during sound wave propagation. For each sound pressure receiving sensor, the geometric path length between the sound pressure receiving sensor and the geometric center of all voxel units is summarized. The sum of all geometric path lengths is divided by the total number of voxel units to obtain the average path length from the sound pressure receiving sensor to the geometric center of all voxel units. The calculated average path length is recorded as the average path length index of the corresponding sound pressure receiving sensor. The average path length of all sound pressure receiving sensors is arithmetically averaged to obtain the overall average path length of all sound pressure receiving sensors and all voxel units within the detection area. Based on the overall average path length index and the total number of voxel units, a uniform initial value is set for the impedance equivalent path area contribution of all voxel units. The initial value is proportional to the square of the overall average path length and inversely proportional to the total number of voxel units, so that all voxel units have the same impedance equivalent path area contribution in the initial state. A mapping relationship is established between the initial values ​​of the equivalent path area contribution of the impedance of all voxel elements and their corresponding voxel numbers, forming a set of variables for sparse reconstruction solution.

[0012] Optionally, the step of constructing an overall optimization objective function based on residual minimization and sparsity constraints, and implementing sparse reconstruction using the equivalent propagation path difference matrix and propagation contribution function to obtain an estimate of the equivalent path area contribution of each voxel element impedance, specifically includes: Based on the propagation contribution function and the equivalent path area contribution of each voxel unit impedance, for each pair of sound pressure receiving sensors, the propagation contribution weight of all voxel units corresponding to each pair of sound pressure receiving sensors is multiplied one by one by the equivalent path area contribution of voxel unit impedance and summed over all voxel units to obtain the theoretical equivalent propagation path difference of the corresponding sensor pair. For each pair of sound pressure receiving sensors, the actual equivalent propagation path difference obtained by cross-correlation analysis is subtracted from the theoretical equivalent propagation path difference calculated by the propagation contribution function and the impedance equivalent path area contribution to obtain the path difference residual of the corresponding sensor pair. For each sound pressure receiving sensor, the sound pressure amplitude of each sampling point in the discrete sound pressure sampling sequence is squared and summed along the entire acquisition period to obtain the sound pressure energy index of the sound pressure receiving sensor during the acquisition period. The sound pressure energy indices of all sound pressure receiving sensors are compared, and the minimum and maximum sound pressure energy indices are recorded respectively. A sparse regularized intensity parameter is constructed based on the ratio of the minimum to the maximum sound pressure energy indices, and the sparse regularized intensity parameter is used as the weight parameter of the sparse regularization term. Under the premise that the contribution of the equivalent path area of ​​impedance of each voxel element is greater than zero, a fractal sparsity function for the contribution of the equivalent path area of ​​impedance of each voxel element is constructed in combination with the overall average path length index. The fractal sparsity function performs a nonlinear transformation on the value of the contribution of the equivalent path area of ​​impedance. The value of the fractal sparsity function increases monotonically with the value of the contribution of the equivalent path area of ​​impedance, and the value of the fractal sparsity function is used as the penalty for the sparsity regularization term. The sum of squares of the path difference residuals of all sensors is used as the data fitting term, and the sum of the fractal sparsity functions of all voxel units after weighting by the sparsity regularization intensity parameter is used as the sparsity regularization term. The two terms are added together to construct the overall optimization objective function. The overall optimization objective function controls the sparse distribution of the contribution of the voxel impedance equivalent path area while maintaining the consistency between the model prediction and the observation data. The partial derivatives of the overall optimization objective function are calculated for the equivalent path area contribution of the impedance of each voxel element. Based on the partial derivative results, an update relationship or iterative solution criterion for the equivalent path area contribution of the impedance of the voxel element is constructed. The overall optimization objective function is solved by numerical optimization methods to obtain the sparse reconstruction results of the equivalent path area contribution of the impedance of each voxel element.

[0013] Optionally, the step of statistically analyzing the impedance equivalent path area contribution of each voxel element, calculating the mean and average absolute deviation index, and identifying abnormal voxel elements according to the degree of deviation specifically includes: After completing the sparse reconstruction and obtaining the equivalent path area contribution of impedance of all voxel elements, the arithmetic mean of the equivalent path area contribution of impedance of all voxel elements is calculated to obtain the mean value of the equivalent path area contribution of impedance of voxel elements. Using the average value of the equivalent path area contribution of the voxel element impedance as a reference, the absolute deviation between the equivalent path area contribution of the voxel element impedance and the average value is calculated for each voxel element. The average absolute deviation index is obtained by averaging the absolute deviations of all voxel elements. For each voxel element, if the calculated absolute deviation is greater than the mean absolute deviation index, it is identified as an abnormal impedance voxel element and marked as a potential defect element. When at least one voxel element in the detection area is identified as an abnormal impedance voxel element, it is determined that there is an internal defect in the concrete within the detection area; when the absolute deviation of all voxel elements does not exceed the average absolute deviation index, it is determined that no significant acoustic anomaly is found within the current detection area, and no obvious internal defect is detected.

[0014] Optionally, the step of clustering anomalous voxel units based on geometric proximity to form defect regions, and outputting their numbers, voxel unit number sets, and average mutation intensity, specifically includes: For all voxel numbers identified as abnormal impedance voxel units, pair them together and calculate the spatial distance between the geometric centers of any two abnormal voxel units. Compare the calculated spatial distance with the length of the voxel diagonal formed by the center spacing of the voxel units in the three coordinate axis directions. When the spatial distance is not greater than the length of the voxel diagonal, the two abnormal voxel units are identified as geometrically adjacent voxel units. By passing on geometric proximity relationships, all anomalous voxel units that are directly or indirectly adjacent to each other are grouped and merged. Each set of anomalous voxel units that are interconnected through proximity relationships is regarded as an independent defect region. A unique number is assigned to each defect region, and the set of all voxel unit numbers contained in each defect region is recorded. For each defect region, the number of voxel units in each defect region is counted based on the set of voxel unit numbers contained in each defect region. For each defect region, the absolute deviation between the impedance equivalent path area contribution of each voxel unit contained in each defect region and the mean of all voxel units is summed, and the summation result is divided by the number of voxel units in each defect region to obtain the average mutation intensity of each defect region. When at least one defect region is detected, the output includes the number of each defect region, the corresponding set of voxel unit numbers, and the average mutation intensity of each defect region; when no abnormal impedance voxel units are detected, the output concludes that no significant internal defects were detected.

[0015] The present invention has the following beneficial effects: 1. This scheme proposes a dynamic equivalent sound velocity calculation strategy based on multiple known propagation paths within a well-preserved region in its sound velocity calibration module. Unlike the traditional fixed sound velocity assumption, this scheme selects multiple pairs of calibration paths within a structurally intact region and calculates the equivalent sound velocity of each receiving sensor in real time using the ratio of path length difference to arrival time difference. The arithmetic mean of these ratios is then used to obtain the reference sound velocity for the detection area. This method overcomes the sound velocity variation problems caused by the non-uniformity of concrete materials and the complexity of the structure, improving the accuracy of subsequent time delay conversion and path difference calculation. Simultaneously, the automatic elimination and re-acquisition mechanism for invalid or zero-time-difference paths ensures the reliability of the calibration results.

[0016] 2. This scheme uses only a single instantaneous impact excitation signal in the excitation and acquisition stages, simultaneously acquiring the full-time domain sound pressure response from multiple sensors, effectively simplifying the setup and operation of the excitation device. Compared with traditional methods involving multiple scans or continuous excitation, a single impact can excite sound waves across the entire frequency band, reducing human interference and instrument drift. The design of a unified sampling frequency and fixed acquisition duration ensures the synchronization of data preprocessing and subsequent cross-correlation analysis. Furthermore, by automatically eliminating failed sensors with all-zero sequences and prompting for redeployment, data integrity is guaranteed. This module avoids data defects caused by insufficient excitation energy or missed acquisitions, obtaining complete and stable time-domain sound pressure data with minimal experimental steps, providing high signal-to-noise ratio raw input for cross-correlation delay estimation.

[0017] 3. In the time delay estimation module, this scheme proposes a two-level cross-correlation analysis process based on a discrete time delay candidate set: first, the cross-correlation value of each pair of sensors is calculated at all time delay candidate points; then, the time delay index with the smallest absolute value is selected first from the set of maximum cross-correlation values; and the actual time delay is uniquely determined in the symmetric solution using a non-negativity rule. This strategy innovatively integrates the dual discrimination principles of minimum absolute offset and non-negativity priority, avoiding time delay ambiguity caused by noise-induced multi-peak or symmetric solutions in traditional cross-correlation peak selection. Furthermore, the cross-correlation calculation is completed at only one moment, eliminating the need for multiple iterations at different excitation moments, thus simplifying the algorithm complexity. This method reduces computational load and time delay while ensuring the accuracy of time delay estimation, facilitating rapid localization in engineering diagnostics.

[0018] 4. For voxel modeling, this scheme divides the detection volume into multiple equidistant voxel units in a local 3D Cartesian coordinate system. Combining the geometric path length between the voxel center and the sensor position, a propagation contribution function is defined, with weights calculated as the reciprocal of the total path length or a monotonically inversely proportional function. Unlike directly using linear weights or empirical parameters, this innovation strictly constructs a contribution matrix based on physical geometric characteristics, accurately reflecting the relative contribution of each voxel to the path differences between different sensors. This function requires no manual adjustment of weight coefficients and can adapt to different volume sizes and sensor placements, exhibiting strong versatility and scalability. The accurate construction of the contribution function provides accurate physical priors for subsequent sparse reconstruction, which is beneficial for improving reconstruction quality and defect location resolution.

[0019] 5. In the variable initialization phase, this scheme innovatively proposes setting the initial value of the equivalent path contribution of all voxel units to the square of the overall average path length divided by the total number of voxel units. This strategy uses the overall average path length as a natural scale, unifying the initial value distribution and avoiding the instability caused by random initialization or empirical settings. Compared with traditional methods that require multiple parameter tunings or rely on historical data, this method "determines the initial value once" based on the field geometric parameters, reducing user involvement and ensuring consistent initialization standards across different detection areas. This advantage enables rapid convergence of reconstruction solution variables, reduces the number of iterations, and prevents local optima traps, providing a reliable starting point for efficient sparse optimization.

[0020] 6. The core module of this scheme integrates the overall optimization objective function with dual constraints of residual minimization and sparse regularization, and introduces data-driven sparse regularization intensity parameters and fractal sparsity functions during the construction process. Unlike conventional L1 regularization or fixed intensity parameters, the sparse regularization intensity is dynamically calculated based on the sensor energy ratio, achieving adaptive weighting for weak signal sensors. The fractal sparsity function, through a nonlinear power-law penalty mechanism based on the overall average path length, imposes strong penalties on large numerical contributions and weak penalties on small numerical contributions, further ensuring the extreme sparsity of the reconstructed solution. This innovation effectively balances model fitting and sparse distribution requirements, overcoming the shortcomings of traditional sparse reconstruction in terms of sensitivity to intensity parameters and overfitting, achieving high-contrast imaging and high-precision localization of internal defects.

[0021] 7. After statistically obtaining the contribution of each voxel, this scheme calculates the overall mean and mean absolute deviation, identifies anomalous voxels using a deviation threshold, and then merges adjacent defect voxels one by one based on geometric proximity to generate independent regions. Compared to threshold segmentation or manual annotation, this method combines statistical discrimination with spatial clustering to automatically complete the identification and merging of anomalous units. This not only improves the accuracy of defect localization but also quantifies regional characteristics such as average mutation intensity and outputs structured results. It can effectively identify multiple scattered or adjacent defects and automatically number and extract features from them, reducing the intensity of manual intervention. It is suitable for large-scale rapid detection scenarios and solves the problem of difficulty in simultaneously handling multiple regions or irregular defects in existing technologies.

[0022] 8. The overall solution organically combines multi-sensor acoustic calibration, instantaneous excitation and unified sampling, refined cross-correlation extraction, sparse reconstruction based on geometric contributions, and automated decision clustering to form an end-to-end concrete internal defect detection system. Compared with traditional ultrasonic, radar, and simple time-of-flight methods, this solution has higher imaging resolution, stronger noise resistance, and lower reliance on manual intervention. It can achieve rapid, reliable, and fully automated structural health monitoring, and has significant engineering application value and promising prospects for widespread application. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the process of the present invention.

[0024] Figure 2 This is a schematic diagram of the three-dimensional rectangular coordinate system structure of the present invention.

[0025] In the diagram: 1 - origin, 2 - Shaft, 3- Shaft, 4- Axis, 5 - Detection area. Detailed Implementation

[0026] 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.

[0027] Example, refer to Figure 1 A method for detecting internal defects in concrete structures, comprising: Select a detection area on the surface of the concrete structure, set up an excitation source and no less than two sound pressure receiving sensors, set up a calibrated propagation path in the intact area, record the path length and arrival time, calculate the equivalent sound velocity of each sensor based on the length difference and time difference, and take the average to obtain the reference sound velocity of the detection area. The excitation source is triggered to emit a unit impact excitation signal, and each sensor collects sound pressure according to uniform sampling parameters to form a discrete-time sound pressure sampling sequence, and the failed sensors with all-zero sequences are eliminated; For any two sensor sequences, perform discrete cross-correlation within a preset time delay range to obtain the maximum cross-correlation time delay, convert it into equivalent propagation path difference according to the reference sound speed in the detection area, and obtain the equivalent propagation path difference matrix of all sensor pairs. Establish a local three-dimensional rectangular coordinate system, divide the volume to be detected into multiple voxel units, record the geometric center of the voxel unit and the coordinates of each sensor, and construct the propagation contribution function based on the geometric path length; Based on the propagation contribution function and geometric path distribution, and combined with the overall average path length index, an initial value for the impedance equivalent path area contribution is set for each voxel element, and a variable set is established. The overall optimization objective function is constructed with residual minimization and sparsity constraints. Sparse reconstruction is carried out using the equivalent propagation path difference matrix and propagation contribution function to obtain the estimated equivalent path area contribution of each voxel element impedance. The impedance equivalent path area contribution of each voxel element is statistically analyzed, and the mean and average absolute deviation index are calculated. Abnormal voxel elements are identified according to the degree of deviation. Abnormal voxel units are clustered based on geometric proximity to form defect regions, and the output numbers, voxel unit number sets, and average mutation intensity are generated.

[0028] By deploying excitation sources and multiple sound pressure receiving sensors on the surface of a concrete structure and performing multipath sound velocity calibration in a known intact area, this method first obtains the field reference parameters for sound wave propagation inside the concrete, overcoming the error accumulation problem caused by the traditional assumption of a constant sound velocity. Subsequently, using the clear time-domain impact response generated by a unit impact excitation, discrete-time sound pressure sequences are obtained according to unified sampling parameters, and all-zero failure channels are eliminated to ensure the reliability of subsequent time difference information extraction. Next, by performing point-by-point cross-correlation analysis on the discrete sequences of any two sensors within a preset time delay range, the maximum correlation time delay is accurately extracted. Combined with the field reference sound velocity, the time delay is converted into an equivalent propagation path difference, constructing a path difference matrix covering all sensor pairs, providing complete observation data for imaging reconstruction. Furthermore, the volume to be inspected is divided into voxel units, and the geometric relationship between each voxel and the sensor is recorded to construct a propagation contribution function, achieving an organic combination of spatial geometric information and path difference data. An overall optimization objective function is constructed using residual minimization and sparsity constraints, and this function is used to iteratively solve for the impedance equivalent path area contribution of each voxel, obtaining a high-resolution estimate of the internal impedance distribution. Finally, by identifying anomalous voxel units and forming defect regions through statistical analysis and spatial clustering, the quantitative output of the location, shape, and abrupt change characteristics of internal defects was achieved. This technology effectively solves the problems of difficulty in locating internal defects in concrete, low imaging accuracy, and susceptibility of imaging results to the non-uniformity of sound velocity. It avoids the subjectivity of manual threshold judgment and has improvements in automation, reliability, and resolution compared to existing technologies.

[0029] The process involves selecting a detection area on the concrete structure surface, deploying an excitation source and at least two sound pressure receiving sensors, setting up a calibrated propagation path in an intact area, recording the path length and arrival time, calculating the equivalent sound velocity of each sensor based on the length and time differences, and averaging the results to obtain the reference sound velocity for the detection area. Specifically, this includes: A detection area is selected on the surface of the concrete bridge pier, an excitation source is set up, and more than two sound pressure receiving sensors are arranged at equal intervals along a predetermined direction on the surface of the selected detection area. Each sound pressure receiving sensor is numbered, and the total number of numbers is the total number of sound pressure receiving sensors in the detection area. In the known intact area of ​​the bridge pier, two fixed calibration propagation paths are selected for each sound pressure receiving sensor. The actual physical propagation path lengths of the two calibration propagation paths are obtained, and the arrival time of the sound wave along the two calibration propagation paths to the corresponding sound pressure receiving sensor is recorded. For each sound pressure receiving sensor, the arrival time difference corresponding to the two calibration propagation paths is compared. When the arrival times of the two calibration propagation paths are the same or the time difference is zero, the calibration result is determined to be invalid. Two calibration propagation paths that meet the geometric difference are selected again near the corresponding sound pressure receiving sensor or the sound wave arrival time data is collected again until two calibration propagation paths with different arrival times are obtained. After obtaining two calibration propagation paths with different arrival times, the equivalent sound velocity of the corresponding sound pressure receiving sensor is calculated based on the difference in actual physical length of the two calibration propagation paths and the corresponding difference in arrival time. When the calculation result is not greater than zero, the calibration result is determined to be invalid, and a new calibration propagation path is selected or data is collected again until an equivalent sound velocity greater than zero is obtained. After calculating the equivalent sound velocity for all sound pressure receiving sensors, the arithmetic mean of all equivalent sound velocities is taken to obtain the reference sound velocity for the detection area.

[0030] Further specific implementation steps include: A detection area was selected on the surface of the bridge pier, and an excitation source was set up with the number [number missing]. And evenly spaced on the surface of this area A sound pressure receiving sensor, numbered as follows: ;in, This refers to the total number of sound pressure receiving sensors deployed within the detection area on the bridge pier surface. The number of the stimulus source; This is the serial number of the sound pressure receiving sensor; Two pairs of fixed path segments are selected within the known intact area of ​​the bridge pier, and their actual physical propagation path lengths are respectively... , The corresponding sound wave reaches the sensor The arrival times are respectively , ;in, , For the first The actual propagation path length of each receiving sensor under the first and second calibrated propagation paths; , The sound waves propagate along the first and second calibration paths to the... The arrival time of each receiving sensor; Recorded , Next, check the arrival time difference: like If the current calibration result is invalid, it is necessary to reselect one of the two calibration paths for the sensor or re-acquire data until the desired result is achieved. ; In satisfying Given the premise that the length difference and arrival time difference of the two paths are the basis for calculating the first path, the second path is calculated. Equivalent speed of sound at each receiving point: ;in, To perform calibration based on two paths for the first The equivalent speed of sound estimated by a single receiving sensor; If the calculation yields Not satisfied If the calibration fails, the calibration is deemed invalid, and a new calibration path or data re-collection is required until the data is obtained. The result; The average of all sound velocity values ​​is taken as the reference sound velocity for the detection area. .

[0031] The triggering excitation source emits a unit impact excitation signal, and each sensor collects sound pressure according to uniform sampling parameters to form a discrete-time sound pressure sampling sequence. Failed sensors with all-zero sequences are eliminated, specifically including: At a preset initial moment, the excitation source is triggered to input a unit impact excitation signal into the concrete structure. The excitation signal has a non-zero amplitude only at the initial moment in the time domain, and the amplitude is zero at other moments. Each sound pressure receiving sensor within the detection area continuously acquires sound pressure response signals from the moment of excitation triggering. During the entire acquisition period, the continuous time function of the sound pressure change of the sound pressure receiving sensor with time is recorded, which constitutes the time domain sound pressure response of each sound pressure receiving sensor. For all sound pressure receiving sensors, a unified sampling frequency and total acquisition time are set. Based on the unified sampling frequency, a discrete sampling time sequence with equal intervals is generated within the acquisition time. The continuous-time sound pressure response of each sound pressure receiving sensor is sampled at each discrete sampling time to form a discrete-time sound pressure sampling sequence for each sound pressure receiving sensor, and the total number of discrete sampling points contained in each sequence is determined. The discrete sampling sequences of all sound pressure receiving sensors are checked one by one. When the sound pressure sampling value of a certain sound pressure receiving sensor is zero at all discrete sampling times, the corresponding sound pressure receiving sensor is determined to be a failed sensor. After redeployment or replacement, the excitation and acquisition operation is performed again until all sound pressure receiving sensors have at least one non-zero sound pressure sampling point during the acquisition period.

[0032] Further specific implementation steps include: At any moment From the source of motivation Send a unit impact signal: ;in, For continuous time variables; For the excitation source in continuous time The output signal strength is below; Each receiving sensor Collect all the sound pressure signals it receives and form a timing function. ;in, For the first Each receiving sensor in continuous time The sound pressure signal measured at any given time; Set a uniform sampling frequency Total collection time Generate a sequence of sampling points: , , ;in, For the index of discrete sampling points; For the first The continuous time values ​​corresponding to each discrete sampling moment; This represents the total number of discrete sampling points. If the data collected by a certain sensor is always zero, that is: , If a sensor fails to register a zero sampling point, it is considered a faulty sensor and needs to be redeployed until all sensors have at least one non-zero sampling point during the data collection period.

[0033] The step involves performing discrete cross-correlation on any two sensor sequences within a preset time delay range to obtain the maximum cross-correlation time delay. This time delay is then converted into an equivalent propagation path difference based on the reference sound velocity in the detection area, resulting in the equivalent propagation path difference matrix for all sensor pairs. Specifically, this includes: Within the maximum analysis delay range determined by a unified sampling frequency, a discrete delay sequence is constructed for any two sound pressure receiving sensors with different numbers. The discrete delay sequence consists of a set of delay candidate values ​​with the sampling time interval as the step size, and the delay index covers the range from the maximum negative delay to the maximum positive delay. For each pair of sound pressure receiving sensors, a candidate time delay is selected from the discrete time delay sequence. One of the sound pressure sampling sequences is time-shifted relative to the other sound pressure sampling sequence under the selected candidate time delay. The results are multiplied point by point at all multiplicable sampling points and accumulated to obtain the discrete cross-correlation value under the selected candidate time delay. The time shift, point-by-point multiplication and accumulation operations are performed on all candidate time delays of each pair of sound pressure receiving sensors to form a sequence of discrete cross-correlation functions with respect to time delay. For each pair of sound pressure receiving sensors, the maximum value of the cross-correlation value is found from the discrete cross-correlation function sequence. Among all the candidate delays that obtain the maximum cross-correlation value, the candidate delay with the smallest absolute value is selected first. When there are two symmetrical candidate delays, the candidate delay with a time value not less than zero is selected from these two candidate delays. The selected candidate delay is taken as the maximum correlation delay of the corresponding sensor pair. For each pair of sound pressure receiving sensors, the maximum correlation delay is converted into the sound wave propagation path difference using the reference sound velocity in the detection area and the corresponding maximum correlation delay, forming the equivalent propagation path difference data of the corresponding sensor pair; the same conversion process is performed on all remaining sensor pairs to obtain the equivalent propagation path difference matrix of all sensor pairs.

[0034] Further specific implementation steps include: For any two receiving sensors and Set the discrete time delay to ;in, and , This is the number of another receiving sensor; The continuous time value corresponding to the discrete time delay; For delay index; This represents the number of sampling points corresponding to the maximum delay. The cross-correlation function is constructed as follows: ;in, In order to delay Next, the and Discrete cross-correlation values ​​between the sound pressure sequences of individual sensors; Set the maximum relevance value to: ;in, For all delays Above, the first and Intercorrelation of individual sensors The maximum value; In all satisfied From the set of delay indices, first select the absolute value. The smallest one; if there are still two symmetric solutions, then choose the one that satisfies... One of them is taken as the unique maximum correlated delay, and this uniquely determined delay is denoted as ;in, The maximum relevant delay selected under the uniqueness rule; Using the reference speed of sound Convert latency into path difference: ;in, To obtain sensor pairs through cross-correlation analysis The corresponding equivalent propagation path difference.

[0035] Reference Figure 2The establishment of a local three-dimensional Cartesian coordinate system, dividing the volume to be detected into multiple voxel units, recording the geometric center of each voxel unit and the coordinates of each sensor, and constructing a propagation contribution function based on the geometric path length, specifically includes: A local three-dimensional rectangular coordinate system is established in the pier inspection area. Any point on the edge of the inspection area is selected as the origin of the coordinate system, and the three coordinate axes are extended along the width direction, height direction and concrete structure thickness direction of the pier, respectively, to form a spatial reference coordinate system. The area to be detected is divided into multiple voxel units along the three coordinate axes according to a preset center spacing. The center spacing in the three directions is the center spacing along the width direction, the center spacing along the height direction, and the center spacing along the thickness direction, respectively. Each voxel unit is assigned a unique number, and the coordinates of the geometric center of each voxel unit in the three-dimensional coordinate system are recorded. The installation position of each sound pressure receiving sensor in the detection area is mapped to a three-dimensional coordinate system, and the spatial coordinates of each sound pressure receiving sensor are recorded. For each voxel unit, based on the three-dimensional distance relationship between the voxel geometric center and the spatial coordinates of each sound pressure receiving sensor, the geometric path length from the voxel geometric center to each sound pressure receiving sensor is calculated. The geometric path length is the spatial straight-line distance in the three-dimensional rectangular coordinate system. For any pair of sound pressure receiving sensors and any voxel unit, the geometric path lengths from the voxel unit to the two sensors are summed. Based on the total path length, a propagation contribution weight is constructed. By setting the reciprocal of the total path length as the weight value, or setting the function value that is monotonically inversely proportional to the total path length as the weight value, the contribution coefficient of the voxel unit to the corresponding sensor to the propagation path is obtained, forming a propagation contribution function about the voxel number and the sensor pair number.

[0036] Further specific implementation steps include: Establish a local three-dimensional rectangular coordinate system in the pier inspection area. The origin is selected at any point on the edge of the detection area, and the three coordinate axes are along the width, depth and height of the pier, respectively. The area to be detected is divided into Individual element, numbered ;in, This refers to the numbering of the voxel unit; The total number of voxel units; The first The spatial location of each sensor is denoted as: ; The first The geometric center position of an individual element is denoted as: ; Let the center-to-center distances of the voxels in the three directions be respectively , , ;in, , , These represent the adjacent voxel centers at... , , Spacing in the direction; Calculate voxels To the sensor geometric path length : ; For sensor pairs With voxels A local three-dimensional Cartesian coordinate system is established, the volume to be detected is divided into multiple voxel units, the geometric center of each voxel unit and the coordinates of each sensor are recorded, and a propagation contribution function is constructed based on the geometric path length, specifically: ;in, In consideration of voxels At that time, the sensor... The weight of the total path contribution; To from the sensor To voxels The geometric distance.

[0037] Based on the propagation contribution function and geometric path distribution, combined with the overall average path length index, an initial value for the impedance equivalent path area contribution is set for each voxel element, and a variable set is established. Specifically, this includes: A scalar parameter is set for each voxel unit in the detection area, and the set scalar parameter is defined as the impedance equivalent path area contribution of the voxel unit during sound wave propagation. For each sound pressure receiving sensor, the geometric path length between the sound pressure receiving sensor and the geometric center of all voxel units is summarized. The sum of all geometric path lengths is divided by the total number of voxel units to obtain the average path length from the sound pressure receiving sensor to the geometric center of all voxel units. The calculated average path length is recorded as the average path length index of the corresponding sound pressure receiving sensor. The average path length of all sound pressure receiving sensors is arithmetically averaged to obtain the overall average path length of all sound pressure receiving sensors and all voxel units within the detection area. Based on the overall average path length index and the total number of voxel units, a uniform initial value is set for the impedance equivalent path area contribution of all voxel units. The initial value is proportional to the square of the overall average path length and inversely proportional to the total number of voxel units, so that all voxel units have the same impedance equivalent path area contribution in the initial state. A mapping relationship is established between the initial values ​​of the equivalent path area contribution of the impedance of all voxel elements and their corresponding voxel numbers, forming a set of variables for sparse reconstruction solution.

[0038] Further specific implementation steps include: Set the first The equivalent path area contribution of an individual element in sound wave propagation is ; First, calculate the average path length from each sensor to all voxels: ;in, For the first The average geometric path length from each sensor to the center of all voxels; Next, calculate the overall average path length for all sensors: ;in, The overall average path length is calculated by further averaging the path lengths of all sensors and all voxels. Therefore, the initial values ​​for the equivalent path area contribution of all voxels are set as follows: ;in, For the first The initial value of the equivalent path area contribution of individual element units.

[0039] The overall optimization objective function is constructed using residual minimization and sparse constraints. Sparse reconstruction is then performed using the equivalent propagation path difference matrix and propagation contribution function to obtain an estimate of the equivalent path area contribution of each voxel element impedance. Specifically, this includes: Based on the propagation contribution function and the equivalent path area contribution of each voxel unit impedance, for each pair of sound pressure receiving sensors, the propagation contribution weight of all voxel units corresponding to each pair of sound pressure receiving sensors is multiplied one by one by the equivalent path area contribution of voxel unit impedance and summed over all voxel units to obtain the theoretical equivalent propagation path difference of the corresponding sensor pair. For each pair of sound pressure receiving sensors, the actual equivalent propagation path difference obtained by cross-correlation analysis is subtracted from the theoretical equivalent propagation path difference calculated by the propagation contribution function and the impedance equivalent path area contribution to obtain the path difference residual of the corresponding sensor pair. For each sound pressure receiving sensor, the sound pressure amplitude of each sampling point in the discrete sound pressure sampling sequence is squared and summed along the entire acquisition period to obtain the sound pressure energy index of the sound pressure receiving sensor during the acquisition period. The sound pressure energy indices of all sound pressure receiving sensors are compared, and the minimum and maximum sound pressure energy indices are recorded respectively. A sparse regularized intensity parameter is constructed based on the ratio of the minimum to the maximum sound pressure energy indices, and the sparse regularized intensity parameter is used as the weight parameter of the sparse regularization term. Under the premise that the contribution of the equivalent path area of ​​impedance of each voxel element is greater than zero, a fractal sparsity function for the contribution of the equivalent path area of ​​impedance of each voxel element is constructed in combination with the overall average path length index. The fractal sparsity function performs a nonlinear transformation on the value of the contribution of the equivalent path area of ​​impedance. The value of the fractal sparsity function increases monotonically with the value of the contribution of the equivalent path area of ​​impedance, and the value of the fractal sparsity function is used as the penalty for the sparsity regularization term. The sum of squares of the path difference residuals of all sensors is used as the data fitting term, and the sum of the fractal sparsity functions of all voxel units after weighting by the sparsity regularization intensity parameter is used as the sparsity regularization term. The two terms are added together to construct the overall optimization objective function. The overall optimization objective function controls the sparse distribution of the contribution of the voxel impedance equivalent path area while maintaining the consistency between the model prediction and the observation data. The partial derivatives of the overall optimization objective function are calculated for the equivalent path area contribution of the impedance of each voxel element. Based on the partial derivative results, an update relationship or iterative solution criterion for the equivalent path area contribution of the impedance of the voxel element is constructed. The overall optimization objective function is solved by numerical optimization methods to obtain the sparse reconstruction results of the equivalent path area contribution of the impedance of each voxel element.

[0040] Further specific implementation steps include: Calculate the theoretical path difference based on the propagation contribution function and the equivalent path area contribution: ;in, For sensor pairs obtained through model calculation The theoretical path is different; Calculate the residual between the actual measured path difference and the theoretical value: ;in, For sensor pair Path difference residual; The data-driven calculation of the sparse canonical strength parameter is as follows: S601. First, calculate the energy index of each sensor: ;in, For the first The energy index of the sound pressure signal from each sensor throughout the entire acquisition period; S602. Calculate the minimum and maximum energy: , ;in, , These are the minimum and maximum values ​​obtained among all sensor energy indicators, respectively; S603, thereby constructing the coefficient canonical intensity parameter ; exist Under the premise of using the overall average path length Construct the fractal sparsity function: ;in, In order to target the Fractal sparsity regularization term of individual elements; It is the power of the overall average path length; For larger Apply non-linear penalties; Based on this, the overall optimization goal is constructed as follows: ;in, To optimize the overall objective function; For each Taking the partial derivative, we get: .

[0041] The process of statistically analyzing the impedance equivalent path area contribution of each voxel element, calculating the mean and average absolute deviation index, and identifying abnormal voxel elements based on the degree of deviation specifically includes: After completing the sparse reconstruction and obtaining the equivalent path area contribution of impedance of all voxel elements, the arithmetic mean of the equivalent path area contribution of impedance of all voxel elements is calculated to obtain the mean value of the equivalent path area contribution of impedance of voxel elements. Using the average value of the equivalent path area contribution of the voxel element impedance as a reference, the absolute deviation between the equivalent path area contribution of the voxel element impedance and the average value is calculated for each voxel element. The average absolute deviation index is obtained by averaging the absolute deviations of all voxel elements. For each voxel element, if the calculated absolute deviation is greater than the mean absolute deviation index, it is identified as an abnormal impedance voxel element and marked as a potential defect element. When at least one voxel element in the detection area is identified as an abnormal impedance voxel element, it is determined that there is an internal defect in the concrete within the detection area; when the absolute deviation of all voxel elements does not exceed the average absolute deviation index, it is determined that no significant acoustic anomaly is found within the current detection area, and no obvious internal defect is detected.

[0042] Further specific implementation steps include: Calculate the mean and mean absolute deviation of the equivalent path area contribution respectively: , ;in, The arithmetic mean of the equivalent path area contribution of all voxels; Contribution of equivalent path area to the mean for all voxels The mean absolute deviation; If a voxel satisfies: If so, it is determined to be an abnormal impedance voxel and marked as a potential defect element; If for all All of the following are available: If no significant acoustic abnormality is found within the current detection area, it is determined that there are no obvious internal defects.

[0043] The process of clustering anomalous voxel units based on geometric proximity to form defect regions, and outputting their numbers, voxel unit number sets, and average mutation intensity, specifically includes: For all voxel numbers identified as abnormal impedance voxel units, pair them together and calculate the spatial distance between the geometric centers of any two abnormal voxel units. Compare the calculated spatial distance with the length of the voxel diagonal formed by the center spacing of the voxel units in the three coordinate axis directions. When the spatial distance is not greater than the length of the voxel diagonal, the two abnormal voxel units are identified as geometrically adjacent voxel units. By passing on geometric proximity relationships, all anomalous voxel units that are directly or indirectly adjacent to each other are grouped and merged. Each set of anomalous voxel units that are interconnected through proximity relationships is regarded as an independent defect region. A unique number is assigned to each defect region, and the set of all voxel unit numbers contained in each defect region is recorded. For each defect region, the number of voxel units in each defect region is counted based on the set of voxel unit numbers contained in each defect region. For each defect region, the absolute deviation between the impedance equivalent path area contribution of each voxel unit contained in each defect region and the mean of all voxel units is summed, and the summation result is divided by the number of voxel units in each defect region to obtain the average mutation intensity of each defect region. When at least one defect region is detected, the output includes the number of each defect region, the corresponding set of voxel unit numbers, and the average mutation intensity of each defect region; when no abnormal impedance voxel units are detected, the output concludes that no significant internal defects were detected.

[0044] Further specific implementation steps include: All voxels that meet the defect conditions are grouped according to their geometric proximity, specifically: S801, Calculating Voxels With voxels Geometric center distance : ; S802, if satisfied If so, it is determined to be a voxel. With voxels Adjacent; By utilizing the transitive closure of adjacency relationships, all interconnected sets of defect voxels are considered as a single defect region, and the first... The set of voxels contained in a defect region is denoted as ,and ;in, This refers to the defect area number; The total number of defective areas detected; For any defect region ,set up For set The number of mid-voxel numbers; Calculate the first Average mutation intensity of the region ; When a defect is detected, the following data is output: Defect area number The set of voxel numbers contained therein Regional mutation intensity ; If no defect is detected, the output will conclude that no significant internal defect was detected.

[0045] 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.

[0046] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting internal defects in concrete structures, characterized in that, include: Select a detection area on the surface of the concrete structure, set up an excitation source and no less than two sound pressure receiving sensors, set up a calibrated propagation path in the intact area, record the path length and arrival time, calculate the equivalent sound velocity of each sensor based on the length difference and time difference, and take the average to obtain the reference sound velocity of the detection area. The excitation source is triggered to emit a unit impact excitation signal, and each sensor collects sound pressure according to uniform sampling parameters to form a discrete-time sound pressure sampling sequence, and the failed sensors with all-zero sequences are eliminated; For any two sensor sequences, perform discrete cross-correlation within a preset time delay range to obtain the maximum cross-correlation time delay, convert it into equivalent propagation path difference according to the reference sound speed in the detection area, and obtain the equivalent propagation path difference matrix of all sensor pairs. Establish a local three-dimensional rectangular coordinate system, divide the volume to be detected into multiple voxel units, record the geometric center of the voxel unit and the coordinates of each sensor, and construct the propagation contribution function based on the geometric path length; Based on the propagation contribution function and geometric path distribution, and combined with the overall average path length index, an initial value for the impedance equivalent path area contribution is set for each voxel element, and a variable set is established. The overall optimization objective function is constructed with residual minimization and sparsity constraints. Sparse reconstruction is carried out using the equivalent propagation path difference matrix and propagation contribution function to obtain the estimated equivalent path area contribution of each voxel element impedance. The impedance equivalent path area contribution of each voxel element is statistically analyzed, and the mean and average absolute deviation index are calculated. Abnormal voxel elements are identified according to the degree of deviation. Abnormal voxel units are clustered based on geometric proximity to form defect regions, and the output numbers, voxel unit number sets, and average mutation intensity are generated.

2. The method for detecting internal defects in concrete structures according to claim 1, characterized in that, The process involves selecting a detection area on the concrete structure surface, deploying an excitation source and at least two sound pressure receiving sensors, setting up a calibrated propagation path in an intact area, recording the path length and arrival time, calculating the equivalent sound velocity of each sensor based on the length and time differences, and averaging the results to obtain the reference sound velocity for the detection area. Specifically, this includes: A detection area is selected on the surface of the concrete bridge pier, an excitation source is set up, and more than two sound pressure receiving sensors are arranged at equal intervals along a predetermined direction on the surface of the selected detection area. Each sound pressure receiving sensor is numbered, and the total number of numbers is the total number of sound pressure receiving sensors in the detection area. In the known intact area of ​​the bridge pier, two fixed calibration propagation paths are selected for each sound pressure receiving sensor. The actual physical propagation path lengths of the two calibration propagation paths are obtained, and the arrival time of the sound wave along the two calibration propagation paths to the corresponding sound pressure receiving sensor is recorded. For each sound pressure receiving sensor, the arrival time difference corresponding to the two calibration propagation paths is compared. When the arrival times of the two calibration propagation paths are the same or the time difference is zero, the calibration result is determined to be invalid. Two calibration propagation paths that meet the geometric difference are selected again near the corresponding sound pressure receiving sensor or the sound wave arrival time data is collected again until two calibration propagation paths with different arrival times are obtained. After obtaining two calibration propagation paths with different arrival times, the equivalent sound velocity of the corresponding sound pressure receiving sensor is calculated based on the difference in actual physical length of the two calibration propagation paths and the corresponding difference in arrival time. When the calculation result is not greater than zero, the calibration result is determined to be invalid, and a new calibration propagation path is selected or data is collected again until an equivalent sound velocity greater than zero is obtained. After calculating the equivalent sound velocity for all sound pressure receiving sensors, the arithmetic mean of all equivalent sound velocities is taken to obtain the reference sound velocity for the detection area.

3. The method for detecting internal defects in concrete structures according to claim 2, characterized in that, The triggering excitation source emits a unit impact excitation signal, and each sensor collects sound pressure according to uniform sampling parameters to form a discrete-time sound pressure sampling sequence. Failed sensors with all-zero sequences are eliminated, specifically including: At a preset initial moment, the excitation source is triggered to input a unit impact excitation signal into the concrete structure. The excitation signal has a non-zero amplitude only at the initial moment in the time domain, and the amplitude is zero at other moments. Each sound pressure receiving sensor within the detection area continuously acquires sound pressure response signals from the moment of excitation triggering. During the entire acquisition period, the continuous time function of the sound pressure change of the sound pressure receiving sensor with time is recorded, which constitutes the time domain sound pressure response of each sound pressure receiving sensor. For all sound pressure receiving sensors, a unified sampling frequency and total acquisition time are set. Based on the unified sampling frequency, a discrete sampling time sequence with equal intervals is generated within the acquisition time. The continuous-time sound pressure response of each sound pressure receiving sensor is sampled at each discrete sampling time to form a discrete-time sound pressure sampling sequence for each sound pressure receiving sensor, and the total number of discrete sampling points contained in each sequence is determined. The discrete sampling sequences of all sound pressure receiving sensors are checked one by one. When the sound pressure sampling value of a certain sound pressure receiving sensor is zero at all discrete sampling times, the corresponding sound pressure receiving sensor is determined to be a failed sensor. After redeployment or replacement, the excitation and acquisition operation is performed again until all sound pressure receiving sensors have at least one non-zero sound pressure sampling point during the acquisition period.

4. The method for detecting internal defects in a concrete structure according to claim 3, characterized in that, The step involves performing discrete cross-correlation on any two sensor sequences within a preset time delay range to obtain the maximum cross-correlation time delay. This time delay is then converted into an equivalent propagation path difference based on the reference sound velocity in the detection area, resulting in the equivalent propagation path difference matrix for all sensor pairs. Specifically, this includes: Within the maximum analysis delay range determined by a unified sampling frequency, a discrete delay sequence is constructed for any two sound pressure receiving sensors with different numbers. The discrete delay sequence consists of a set of delay candidate values ​​with the sampling time interval as the step size, and the delay index covers the range from the maximum negative delay to the maximum positive delay. For each pair of sound pressure receiving sensors, a candidate time delay is selected from the discrete time delay sequence. One of the sound pressure sampling sequences is time-shifted relative to the other sound pressure sampling sequence under the selected candidate time delay. The results are multiplied point by point at all multiplicable sampling points and accumulated to obtain the discrete cross-correlation value under the selected candidate time delay. The time shift, point-by-point multiplication and accumulation operations are performed on all candidate time delays of each pair of sound pressure receiving sensors to form a sequence of discrete cross-correlation functions with respect to time delay. For each pair of sound pressure receiving sensors, the maximum value of the cross-correlation value is found from the discrete cross-correlation function sequence. Among all the candidate delays that obtain the maximum cross-correlation value, the candidate delay with the smallest absolute value is selected first. When there are two symmetrical candidate delays, the candidate delay with a time value not less than zero is selected from these two candidate delays. The selected candidate delay is taken as the maximum correlation delay of the corresponding sensor pair. For each pair of sound pressure receiving sensors, the maximum correlation delay is converted into the sound wave propagation path difference using the reference sound velocity in the detection area and the corresponding maximum correlation delay, forming the equivalent propagation path difference data of the corresponding sensor pair; the same conversion process is performed on all remaining sensor pairs to obtain the equivalent propagation path difference matrix of all sensor pairs.

5. The method for detecting internal defects in a concrete structure according to claim 4, characterized in that, The process of establishing a local three-dimensional Cartesian coordinate system, dividing the volume to be detected into multiple voxel units, recording the geometric center of each voxel unit and the coordinates of each sensor, and constructing a propagation contribution function based on the geometric path length specifically includes: A local three-dimensional rectangular coordinate system is established in the pier inspection area. Any point on the edge of the inspection area is selected as the origin of the coordinate system, and the three coordinate axes are extended along the width direction, height direction and concrete structure thickness direction of the pier, respectively, to form a spatial reference coordinate system. The area to be detected is divided into multiple voxel units along the three coordinate axes according to a preset center spacing. The center spacing in the three directions is the center spacing along the width direction, the center spacing along the height direction, and the center spacing along the thickness direction, respectively. Each voxel unit is assigned a unique number, and the coordinates of the geometric center of each voxel unit in the three-dimensional coordinate system are recorded. The installation position of each sound pressure receiving sensor in the detection area is mapped to a three-dimensional coordinate system, and the spatial coordinates of each sound pressure receiving sensor are recorded. For each voxel unit, based on the three-dimensional distance relationship between the voxel geometric center and the spatial coordinates of each sound pressure receiving sensor, the geometric path length from the voxel geometric center to each sound pressure receiving sensor is calculated. The geometric path length is the spatial straight-line distance in the three-dimensional rectangular coordinate system. For any pair of sound pressure receiving sensors and any voxel unit, the geometric path lengths from the voxel unit to the two sensors are summed. Based on the total path length, a propagation contribution weight is constructed. By setting the reciprocal of the total path length as the weight value, or setting the function value that is monotonically inversely proportional to the total path length as the weight value, the contribution coefficient of the voxel unit to the corresponding sensor to the propagation path is obtained, forming a propagation contribution function about the voxel number and the sensor pair number.

6. The method for detecting internal defects in a concrete structure according to claim 5, characterized in that, Based on the propagation contribution function and geometric path distribution, combined with the overall average path length index, an initial value for the impedance equivalent path area contribution is set for each voxel element, and a variable set is established. Specifically, this includes: A scalar parameter is set for each voxel unit in the detection area, and the set scalar parameter is defined as the impedance equivalent path area contribution of the voxel unit during sound wave propagation. For each sound pressure receiving sensor, the geometric path length between the sound pressure receiving sensor and the geometric center of all voxel units is summarized. The sum of all geometric path lengths is divided by the total number of voxel units to obtain the average path length from the sound pressure receiving sensor to the geometric center of all voxel units. The calculated average path length is recorded as the average path length index of the corresponding sound pressure receiving sensor. The average path length of all sound pressure receiving sensors is arithmetically averaged to obtain the overall average path length of all sound pressure receiving sensors and all voxel units within the detection area. Based on the overall average path length index and the total number of voxel units, a uniform initial value is set for the impedance equivalent path area contribution of all voxel units. The initial value is proportional to the square of the overall average path length and inversely proportional to the total number of voxel units, so that all voxel units have the same impedance equivalent path area contribution in the initial state. A mapping relationship is established between the initial values ​​of the equivalent path area contribution of the impedance of all voxel elements and their corresponding voxel numbers, forming a set of variables for sparse reconstruction solution.

7. The method for detecting internal defects in a concrete structure according to claim 6, characterized in that, The overall optimization objective function is constructed using residual minimization and sparse constraints. Sparse reconstruction is then performed using the equivalent propagation path difference matrix and propagation contribution function to obtain an estimate of the equivalent path area contribution of each voxel element impedance. Specifically, this includes: Based on the propagation contribution function and the equivalent path area contribution of each voxel unit impedance, for each pair of sound pressure receiving sensors, the propagation contribution weight of all voxel units corresponding to each pair of sound pressure receiving sensors is multiplied one by one by the equivalent path area contribution of voxel unit impedance and summed over all voxel units to obtain the theoretical equivalent propagation path difference of the corresponding sensor pair. For each pair of sound pressure receiving sensors, the actual equivalent propagation path difference obtained by cross-correlation analysis is subtracted from the theoretical equivalent propagation path difference calculated by the propagation contribution function and the impedance equivalent path area contribution to obtain the path difference residual of the corresponding sensor pair. For each sound pressure receiving sensor, the sound pressure amplitude of each sampling point in the discrete sound pressure sampling sequence is squared and summed along the entire acquisition period to obtain the sound pressure energy index of the sound pressure receiving sensor during the acquisition period. The sound pressure energy indices of all sound pressure receiving sensors are compared, and the minimum and maximum sound pressure energy indices are recorded respectively. A sparse regularized intensity parameter is constructed based on the ratio of the minimum to the maximum sound pressure energy indices, and the sparse regularized intensity parameter is used as the weight parameter of the sparse regularization term. Under the premise that the contribution of the equivalent path area of ​​impedance of each voxel element is greater than zero, a fractal sparsity function for the contribution of the equivalent path area of ​​impedance of each voxel element is constructed in combination with the overall average path length index. The fractal sparsity function performs a nonlinear transformation on the value of the contribution of the equivalent path area of ​​impedance. The value of the fractal sparsity function increases monotonically with the value of the contribution of the equivalent path area of ​​impedance, and the value of the fractal sparsity function is used as the penalty for the sparsity regularization term. The sum of squares of the path difference residuals of all sensors is used as the data fitting term, and the sum of the fractal sparsity functions of all voxel units after weighting by the sparsity regularization intensity parameter is used as the sparsity regularization term. The two terms are added together to construct the overall optimization objective function. The overall optimization objective function controls the sparse distribution of the contribution of the voxel impedance equivalent path area while maintaining the consistency between the model prediction and the observation data. The partial derivatives of the overall optimization objective function are calculated for the equivalent path area contribution of the impedance of each voxel element. Based on the partial derivative results, an update relationship or iterative solution criterion for the equivalent path area contribution of the impedance of the voxel element is constructed. The overall optimization objective function is solved by numerical optimization methods to obtain the sparse reconstruction results of the equivalent path area contribution of the impedance of each voxel element.

8. The method for detecting internal defects in a concrete structure according to claim 7, characterized in that, The process of statistically analyzing the impedance equivalent path area contribution of each voxel element, calculating the mean and average absolute deviation index, and identifying abnormal voxel elements based on the degree of deviation specifically includes: After completing the sparse reconstruction and obtaining the equivalent path area contribution of impedance of all voxel elements, the arithmetic mean of the equivalent path area contribution of impedance of all voxel elements is calculated to obtain the mean value of the equivalent path area contribution of impedance of voxel elements. Using the average value of the equivalent path area contribution of the voxel element impedance as a reference, the absolute deviation between the equivalent path area contribution of the voxel element impedance and the average value is calculated for each voxel element. The average absolute deviation index is obtained by averaging the absolute deviations of all voxel elements. For each voxel element, if the calculated absolute deviation is greater than the mean absolute deviation index, it is identified as an abnormal impedance voxel element and marked as a potential defect element. When at least one voxel element in the detection area is identified as an abnormal impedance voxel element, it is determined that there is an internal defect in the concrete within the detection area; when the absolute deviation of all voxel elements does not exceed the average absolute deviation index, it is determined that no significant acoustic anomaly is found within the current detection area, and no obvious internal defect is detected.

9. A method for detecting internal defects in a concrete structure according to claim 8, characterized in that, The process of clustering anomalous voxel units based on geometric proximity to form defect regions, and outputting their numbers, voxel unit number sets, and average mutation intensity, specifically includes: For all voxel numbers identified as abnormal impedance voxel units, pair them together and calculate the spatial distance between the geometric centers of any two abnormal voxel units. Compare the calculated spatial distance with the length of the voxel diagonal formed by the center spacing of the voxel units in the three coordinate axis directions. When the spatial distance is not greater than the length of the voxel diagonal, the two abnormal voxel units are identified as geometrically adjacent voxel units. By passing on geometric proximity relationships, all anomalous voxel units that are directly or indirectly adjacent to each other are grouped and merged. Each set of anomalous voxel units that are interconnected through proximity relationships is regarded as an independent defect region. A unique number is assigned to each defect region, and the set of all voxel unit numbers contained in each defect region is recorded. For each defect region, the number of voxel units in each defect region is counted based on the set of voxel unit numbers contained in each defect region. For each defect region, the absolute deviation between the impedance equivalent path area contribution of each voxel unit contained in each defect region and the mean of all voxel units is summed, and the summation result is divided by the number of voxel units in each defect region to obtain the average mutation intensity of each defect region. When at least one defect region is detected, the output includes the number of each defect region, the corresponding set of voxel unit numbers, and the average mutation intensity of each defect region; when no abnormal impedance voxel units are detected, the output concludes that no significant internal defects were detected.

Images