Non-destructive testing methods and equipment for deep cracks in road surface repair

Abstract

This invention relates to the field of pavement inspection technology, specifically to a method and equipment for non-destructive testing of deep cracks in pavement repair. The method acquires laser ranging, the first wave travel time of the shallow pavement layer, and the reference sequence number of the laser ranging, binding them to geographic coordinates and the original radar transmission sequence number. Based on the changes in laser ranging and the first wave travel time of the shallow pavement layer, it obtains the elevation-travel time difference ratio vector. Based on the fixed mechanical spacing and sequence number of the sensors, it constructs a pulse spatial offset matrix and maps it to a kernel matrix, which is then alternately scaled and iterated with the two ratio vectors to obtain a pulse probability matrix. This matrix is ​​then used to extract the original ground-penetrating radar waveform data and stitch it together to form a spatiotemporal echo matrix. Based on the changes in laser ranging and the spatiotemporal echo matrix, combined with the reflection travel time, it obtains the location of the defect and its remaining safe distance, thereby generating milling commands. This invention effectively overcomes data misalignment caused by wheel slippage and errors in fixed empirical wave velocity, providing a precise cutting command control closed loop for automatic pavement repair.

Description

Technical Field

[0001] This invention relates to the field of pavement inspection technology, and more specifically to a method and equipment for non-destructive testing of deep cracks for pavement repair. Background Technology

[0002] Deep cracks in municipal road surfaces are a common structural defect in pavement layers. In automated road maintenance and repair operations, the core basis for determining the depth of mechanical milling is the actual depth of the defect extending downwards. Existing non-destructive testing vehicles typically use a front-mounted 3D laser scanner to acquire the geometric undulations of the road surface, and a rear-mounted ground-penetrating radar to acquire the travel time reflected by the internal medium. In areas with limited satellite positioning signals, such as under overpasses or in tunnels, the data alignment of the front and rear heterogeneous sensors relies entirely on the ranging encoder pulses of the chassis wheels.

[0003] However, in practice, when the detection vehicle travels through potholes or waterlogged sections of road, the wheels are prone to localized slippage and spinning, causing the ranging encoder to generate redundant trigger pulses in place. Since there is a fixed physical installation distance between the front laser scanner and the rear ground-penetrating radar, their travel times through the same slippage area differ. These redundant pulses cause a non-linear spatial offset in the time sequence between the laser ranging data and the ground-penetrating radar data, resulting in a severe coordinate mismatch between the surface geometric features of the damage and the underground electromagnetic reflection features. Furthermore, existing technologies for calculating damage depth using ground-penetrating radar echo travel time are flawed. The propagation speed of radar electromagnetic waves is calculated by using a fixed empirical dielectric constant for the entire road section. However, at actual road cracks, the looseness of the material and the infiltration of water are dynamic. The water-rich and loose state will cause unpredictable additional delays in electromagnetic waves. The fixed wave velocity cannot reflect the true electromagnetic delay state of the local disease medium, resulting in serious distortion in the quantitative calculation of the remaining thickness of the underlying disease. This misjudgment of depth ultimately makes it impossible for the control system to issue accurate single-layer or cross-layer cutting control commands to the mechanical milling equipment, which can easily lead to over-milling and waste of materials, or under-milling and recurrence of disease residues. Summary of the Invention

[0004] To address the technical problems of existing non-destructive testing methods being unable to effectively handle nonlinear misalignment of heterogeneous sensor data caused by wheel slippage, and the inaccurate quantification of local damage depth due to the use of fixed empirical wave velocities, thus failing to provide accurate milling cutting commands for road repair equipment, the present invention aims to provide a method and equipment for non-destructive testing of deep cracks in road repair. The specific technical solution adopted is as follows: In a first aspect, one embodiment of the present invention provides a non-destructive testing method for deep cracks in road surface repair, the method comprising the following steps: Obtain the first wave travel time of the shallow surface layer of laser ranging and ground penetrating radar within a single road segment; obtain the reference sequence number of laser ranging and bind it to geographic coordinates; obtain the original radar transmission sequence number; Based on the changes in the first wave travel time of laser ranging and shallow road surface, the elevation difference proportion vector and travel time difference proportion vector within a single road segment are obtained sequentially; based on the mechanical fixed distance between the laser scanner and the ground penetrating radar in the longitudinal direction, the reference number, and the original radar transmission number, a pulse spatial offset matrix is ​​constructed; the pulse spatial offset matrix is ​​mapped to a kernel matrix, and alternately scaled and iterated with the elevation difference proportion vector and the travel time difference proportion vector to obtain the pulse probability matrix; Based on the target column index corresponding to the largest element in each row of the pulse probability matrix, single-column waveform data corresponding to the original three-dimensional echo of the ground penetrating radar are extracted and spliced ​​to generate a spatiotemporal echo matrix; the location of the defect and its local correction coefficient are obtained based on the laser ranging variation combined with the spatiotemporal echo matrix; the remaining safe distance is obtained based on the reflection travel time of the waveform at the defect location and the local correction coefficient. Based on the remaining safety distance and geographic coordinates, generate milling instructions.

[0005] Furthermore, the methods for obtaining the elevation difference proportion vector and the travel time difference proportion vector are as follows: The laser ranging sequences are arranged in the order of acquisition to obtain the laser ranging sequence; the first wave travel times of the shallow road surface are arranged in the order of acquisition to obtain the first wave travel times of the shallow road surface. For any sequence in the laser ranging sequence and the shallow road surface first wave travel time sequence, obtain the first-order absolute difference of adjacent data in the sequence, and use the sum of all first-order absolute differences as the total difference value. The sum of each first-order absolute difference and a preset first positive number is used as the numerator, and the sum of the total difference and a preset second positive number is used as the denominator. Each numerator is divided by the denominator to obtain the difference proportion value and concatenate them into a difference proportion vector. The differential proportion vector corresponding to the laser ranging sequence is used as the elevation differential proportion vector; the differential proportion vector corresponding to the first wave travel time sequence of the shallow road surface is used as the travel time differential proportion vector.

[0006] Furthermore, the method for obtaining the pulse space offset matrix is ​​as follows: Divide the mechanical fixed interval by the nominal mileage of a single pulse to obtain the theoretical number of delayed pulses; For any original radar transmission sequence number and any reference sequence number, the absolute value of the difference between the original radar transmission sequence number and the reference sequence number, and then the theoretical delay pulse number, is used as the pulse space offset penalty value corresponding to the original radar transmission sequence number and the reference sequence number. The pulse space offset matrix is ​​generated by using the reference sequence number as the row of the matrix, the original radar transmission sequence number as the column of the matrix, and the pulse space offset penalty value as the element of the matrix.

[0007] Furthermore, the method for obtaining the pulse probability matrix is ​​as follows: Each element in the pulse space offset matrix is ​​subjected to negative exponential decay to generate a kernel matrix; Initialize the left and right column vectors, multiply the kernel matrix with the right column vector, add each element of the multiplication result to the preset third positive number to generate the first denominator vector, divide each element in the elevation difference proportion vector by the element at the corresponding position in the first denominator vector, and update the left column vector. Multiply the transpose of the kernel matrix with the left column vector, add each element of the multiplication result to the preset fourth positive number to generate the second denominator vector, divide each element in the travel time difference percentage vector by the element at the corresponding position in the second denominator vector, and update the right column vector. Alternately perform update operations on the left and right column vectors until a preset stopping condition is met, and obtain the converged left and right column vectors; Expand the converged left and right column vectors into first and second diagonal matrices, respectively. Multiply the first diagonal matrix, the kernel matrix, and the second diagonal matrix in sequence to obtain the impulse probability matrix.

[0008] Furthermore, the method for obtaining the spatiotemporal echo matrix is ​​as follows: For any row in the pulse probability matrix, the column index corresponding to the largest element in that row is taken as the target column index of that row; if there are at least two column indices corresponding to the largest element in that row, then each column index is subtracted from the corresponding baseline number, and then the theoretical delay pulse number is subtracted to obtain the reference level of the corresponding column index; the column index corresponding to the minimum absolute value of the reference level is taken as the target column index of that row. Extract single-column waveform data corresponding to all target column indices from the raw three-dimensional echo of the ground penetrating radar; All extracted single-column waveform data are concatenated sequentially according to their corresponding reference numbers to generate a spatiotemporal echo matrix.

[0009] Furthermore, the method for obtaining the location of the disease is as follows: Extreme distance points are extracted from the laser ranging sequence, and the geographic coordinates corresponding to the reference number of each extreme distance point are taken as the location of the disease.

[0010] Furthermore, the method for obtaining the local correction coefficients is as follows: For any location of disease, the average value of laser ranging within a preset adjacent range of the extreme ranging point corresponding to the location of disease is used as the reference elevation corresponding to the location of disease. The absolute value of the difference between the extreme distance point corresponding to the location of the disease and the benchmark elevation is taken as the concavity depth of the location of the disease. When the depression depth is less than or equal to a preset depth threshold, the preset wave velocity constant is used as the local correction coefficient for the location of the defect. When the depression depth is greater than a preset depth threshold, extract the single-column waveform data corresponding to the location of the defect in the spatiotemporal echo matrix, and take the absolute value of the difference between the travel time of the first peak signal exceeding the preset background threshold in the single-column waveform data and the reference reflection travel time of the healthy road surface as the shallow damage travel time; wherein, the reference reflection travel time of the healthy road surface is the average value of the first wave travel time of the radar in the undamaged area within a preset range of the location of the defect; The sum of the reference reflection travel time of the healthy pavement and the travel time of the shallow damage is taken as the first result; the reference reflection travel time of the healthy pavement is divided by the first result to obtain the wave velocity attenuation factor; the preset wave velocity constant is multiplied by the wave velocity attenuation factor to obtain the local correction coefficient for the location of the defect.

[0011] Furthermore, the method for obtaining the remaining safe distance is as follows: For any location of a defect, extract the single-column waveform data corresponding to that location from the spatiotemporal echo matrix. Utilize the extreme value mutation characteristics of the reflection envelope amplitude, set a time-domain shielding window starting from the first wave travel time of the shallow road surface layer at that defect location plus a preset blind zone delay. Within the waveform segment range after avoiding the time-domain shielding window, obtain the time corresponding to the first feature point exceeding the warning amplitude threshold from the single-column waveform data as the first reflection travel time corresponding to the bottom of the defect. Within the set adjacent healthy structure layer depth judgment window, obtain the second reflection travel time corresponding to the interface of the adjacent healthy structure layer based on the zero-crossing jump point of the first derivative. Subtract the first reflection time from the second reflection time to obtain the time difference value of the disease location; The remaining safe distance at the location of the defect is obtained by multiplying the time difference at the defect location by the local correction coefficient and dividing by a constant 2.

[0012] Furthermore, the method for generating milling instructions based on the remaining safety distance and geographic coordinates is as follows: For any defect location, when the remaining safe distance of the defect location is greater than or equal to the preset cutting buffer threshold, the preset single-layer cutting depth parameter is spliced ​​with the geographic coordinates corresponding to the defect location to generate a milling command. When the remaining safe distance at the location of the defect is less than the preset cutting buffer threshold, the preset standard milling depth is added to the preset structural layer thickness to generate a cross-layer cutting depth parameter; the cross-layer cutting depth parameter is then combined with the geographic coordinates corresponding to the location of the defect to generate a milling command.

[0013] Secondly, another embodiment of the present invention provides a non-destructive testing device for deep cracks in road surface repair. The device includes: a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above methods.

[0014] The present invention has the following beneficial effects: This invention first obtains the elevation difference proportion vector and travel time difference proportion vector within a single road segment based on the changes in laser ranging and the first wave travel time of the shallow road surface. This facilitates the establishment of a unified dimensionless feature benchmark, enabling similarity alignment calculations for heterogeneous sensor data within the same mathematical framework. To impose physical boundary constraints on the massive redundant pulses generated by slippage, a pulse space offset matrix is ​​constructed based on the longitudinal mechanical fixed distance between the laser scanner and the ground-penetrating radar, the benchmark sequence number, and the original radar transmission sequence number. This matrix accurately reflects the objective punishment of radar's measured triggering timing violating the rigid mechanical distance of the vehicle chassis. The degree of penalty is beneficial for effectively intercepting extremely unreasonable spatial misalignment matching during the optimization process; in order to comprehensively consider the distribution matching degree of road surface abrupt change characteristics and the rigid constraints of vehicle hardware size, the pulse spatial offset matrix is ​​mapped to a kernel matrix, and alternately scaled and iterated with the elevation difference proportion vector and the travel time difference proportion vector to obtain the pulse probability matrix, which accurately reflects the confidence that any original radar transmitted pulse should belong to the corresponding laser reference number, which is beneficial for providing a core allocation mapping reference for the reconstruction of downstream coaxial data; in order to preserve the original appearance of the high-frequency signal used for defect detection and avoid waveform numerical accumulation, The destructive interference of the pulse probability matrix is ​​used to extract single-column waveform data corresponding to the original three-dimensional echo of the ground-penetrating radar based on the target column index corresponding to the largest element in each row. This data is then stitched together to generate a spatiotemporal echo matrix, achieving a one-to-one correspondence between the original data coordinates and facilitating the construction of a fully coaxial spatiotemporal data base. Furthermore, based on the laser ranging variation combined with the spatiotemporal echo matrix, the location of the disease and its local correction coefficients are obtained, determining the specific spatial characteristic coordinates of the area requiring maintenance intervention. This accurately reflects the local equivalent attenuation characteristics of the ground-penetrating radar wave transmission speed caused by the loose material and moisture accumulation state inside the disease point, overcoming the limitations imposed by fixed empirical wave velocities. To accurately quantify the actual depth of the disease extending downwards and define the integrity of the load-bearing structure, the remaining safe distance is obtained based on the reflection travel time of the waveform at the disease location and the local correction coefficient. This accurately reflects the true physical thickness of the interface between the bottom of the disease and the underlying healthy structure, providing a reliable basis for formulating a precise cutting plan to prevent recurrence. Furthermore, based on the remaining safe distance and geographical coordinates, milling instructions are accurately generated to achieve fully automated closed-loop control of the entire process of distance alignment, depth assessment, and milling decision. This effectively eliminates hidden dangers of underlying diseases and avoids excessive milling that wastes materials or insufficient cutting that leads to disease recurrence. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart illustrating a method for non-destructive testing of deep cracks in road surface repair, provided in one embodiment of the present invention. Figure 2 This is a pulse space offset matrix provided in one embodiment of the present invention; Figure 3This is a heatmap of pulse probability matrix calculation under wheel slippage conditions provided in one embodiment of the present invention; Figure 4 This is a comparison chart of milling depth calculation errors provided in one embodiment of the present invention. Detailed Implementation

[0016] The following description, in conjunction with the accompanying drawings, details the specific scheme of the non-destructive testing method and equipment for deep cracks in road surface repair provided by this invention.

[0017] Example 1: This invention proposes a non-destructive testing method for deep cracks in road surface repair. Please refer to [link / reference]. Figure 1 The diagram illustrates a schematic flowchart of a non-destructive testing method for deep cracks in road surface repair according to an embodiment of the present invention. The method includes the following steps: Step S1: Obtain the first wave travel time of the laser ranging and ground penetrating radar in the shallow surface layer within a single road segment; obtain the reference sequence number of the laser ranging and bind it to the geographic coordinates; obtain the original radar transmission sequence number.

[0018] Specifically, existing methods for detecting deep cracks in municipal road surfaces typically utilize vehicle-mounted non-destructive testing (NDT) systems to automatically acquire characteristic data of the road surface and internal media during operation. In these systems, a laser scanner is located at the front of the vehicle's chassis, emitting a laser array downwards and receiving reflected beams to obtain the geometric elevation undulations of the road surface. A ground-penetrating radar (GPR) is located at the rear of the vehicle's chassis, emitting high-frequency electromagnetic waves into the road surface and receiving reflected signals to detect the echo travel time and hidden defects of the internal media. To overcome the nonlinear misalignment of sensor trigger pulses caused by wheel slippage when the vehicle passes over potholes or waterlogged sections, this embodiment analyzes a single road segment as an example. Subsequent instances of "single road segment" refer to the same single road segment; a single road segment is the physical interval of a single continuous driving operation of the vehicle-mounted NDT system.

[0019] Considering the strong co-occurrence of surface geometric elevation data obtained from a front-mounted laser scanner and underground electromagnetic reflection characteristics obtained from a rear-mounted ground-penetrating radar (GPR) at the location of spatial defects, this embodiment establishes a computational basis for aligning misaligned data. This is achieved by continuously acquiring the vertical distance between the scanner and the road surface within a single road segment using a front-mounted 3D laser scanner mounted at the front of the inspection vehicle, serving as laser ranging. Simultaneously, the GPR system mounted at the rear of the inspection vehicle continuously acquires echo signals within a single road segment, and uses a first-derivative peak-finding algorithm to extract the two-way flight time corresponding to the first significant amplitude jump, serving as the shallow first-wave travel time of the GPR. The first-derivative peak-finding algorithm is well-known and will not be elaborated further.

[0020] To create a globally aligned scale unaffected by slippage, and to obtain the reference sequence number for laser ranging and bind it to geographic coordinates, the process involves: capturing all synchronous trigger pulse signals emitted by the chassis hardware encoder within a single road segment; strictly incrementing the accumulated pulse count sequentially to serve as the reference sequence number for laser ranging; and simultaneously reading the geographic mileage marker received by the vehicle control terminal in real time, writing the current reference sequence number and the obtained geographic mileage marker into memory to establish a one-to-one correspondence, thus completing the binding of geographic coordinates. This operation provides a precise physical location basis for the subsequent generation of mechanical milling instructions. Furthermore, considering that the number of pulses emitted by the ground-penetrating radar is often greater when the wheels are slipping and spinning than when the vehicle is in normal driving condition, the original radar transmission sequence number is obtained. Specifically, the triggering actions of the ground-penetrating radar during the actual scanning process are independently counted and recorded in ascending order of triggering sequence to obtain the original radar transmission sequence number. Considering that the subsequent peak-finding operation needs to construct a complete three-dimensional matrix profile based on the continuous transmission time sequence, the original radar transmission sequence number is used as the column index of the original ground-penetrating radar echo data to ensure the complete mapping of electromagnetic reflection characteristics on the time axis and to provide an initial positioning coordinate reference for the downstream time-series alignment calculation.

[0021] Step S2: Based on the changes in the first wave travel time of the laser ranging and the shallow road surface, obtain the elevation difference proportion vector and travel time difference proportion vector within a single road segment in sequence; construct the pulse spatial offset matrix based on the mechanical fixed distance between the laser scanner and the ground penetrating radar in the longitudinal direction, the reference number, and the original radar transmission number; map the pulse spatial offset matrix into a kernel matrix, and perform alternating scaling and iteration with the elevation difference proportion vector and the travel time difference proportion vector to obtain the pulse probability matrix.

[0022] Specifically, considering the inherent differences in physical dimensions and numerical scales between laser ranging (distance dimension, millimeter level) and the first wave travel time of shallow road surface (time dimension, nanosecond level), in order to establish a unified feature benchmark and enable heterogeneous sensor data to be aligned and calculated within the same mathematical framework, and based on the changes in laser ranging and the first wave travel time of shallow road surface, the elevation difference proportion vector and travel time difference proportion vector within a single road segment are obtained sequentially. This accurately characterizes the relative intensity proportion of road surface abrupt changes (geometric indentation and loose internal medium) in the global context of a single road segment, which is beneficial for subsequent nonlinear temporal remapping through the similarity of feature distributions. Furthermore, considering the initial spatiotemporal misalignment of multi-source data—that is, the massive redundant pulses generated by the radar during wheel slippage and the laser reference pulses exhibit severe temporal stacking and deviation—a pulse spatial offset matrix is ​​constructed based on the longitudinal mechanical fixed distance between the laser scanner and the ground-penetrating radar, the reference sequence number, and the original radar transmission sequence number. This matrix accurately reflects the objective penalty degree for the radar's measured triggering sequence violating the chassis's rigid mechanical dimensions, facilitating subsequent interception of extremely unreasonable misaligned matches. To achieve the transformation of the cost matrix from a penalty map to the non-negative kernel function required for iterative solution, the pulse spatial offset matrix is ​​mapped to a kernel matrix, accurately reflecting the matching feasibility kernel value between different pulse combinations. To comprehensively consider the distribution matching degree of road surface abrupt changes and the rigid constraints of vehicle hardware dimensions, the matrix is ​​further iteratively scaled with the elevation difference proportion vector and the travel time difference proportion vector to obtain a pulse probability matrix. This matrix accurately reflects the probability that any original radar transmission pulse should belong to the corresponding laser reference sequence number, providing a core data benchmark for subsequent analysis.

[0023] Preferably, in one feasible manner of this embodiment, the method for obtaining the elevation difference proportion vector and the travel time difference proportion vector is as follows: First, establish a time series basic data segment to facilitate subsequent one-dimensional difference calculation and feature distribution extraction, and then arrange the laser ranging according to the acquisition order to obtain the laser ranging sequence; arrange the first wave travel time of the shallow road surface according to the acquisition order to obtain the first wave travel time sequence of the shallow road surface. Then, in order to extract the local abrupt changes in road surface geometry and internal medium travel time and filter out the globally stationary basis constants, for any sequence in the laser ranging sequence and the first wave travel time sequence of the shallow road surface, the first-order absolute difference between adjacent data in that sequence is obtained. ;in, This refers to the i-th data in the sequence; This refers to the (i-1)th data element in the sequence. The absolute value symbol is used. It should be noted that, to ensure the length of the first-order absolute difference is strictly consistent with the length of the sequence, the first-order absolute difference corresponding to the first data point in the sequence is set to 0 to establish an alignment boundary. To form a base for percentage calculation and accurately reflect the overall cumulative amount of abrupt changes within a single road segment, the sum of all first-order absolute differences is used as the total difference value. Considering that the first-order absolute differences and their total value approach zero in smooth, defect-free road segments, to establish a smooth buffer for mathematical operations, the sum of each first-order absolute difference and a preset first positive number is used as the numerator to avoid extreme feature scaling caused by an absolute zero numerator. The sum of the total difference value and a preset second positive number is used as the denominator to avoid triggering a division-by-zero overflow anomaly when the total difference value is zero in smooth areas. In this embodiment, both the preset first positive number and the preset second positive number are set to be... Furthermore, the units are all consistent with the units of the elements in the sequence, ensuring that the preset first positive number and the preset second positive number will not cause numerical interference to the actual proportion distribution of the true road surface mutation characteristics; the implementer can set the preset first positive number and the preset second positive number according to the data acquisition accuracy of the sensor, and there is no limitation here; Finally, in order to complete the dimensionless feature transformation and map the millimeter-level distance data and nanosecond-level time data to the same value range, each numerator is divided by the denominator to obtain the differential proportion value and concatenate them into a differential proportion vector. Among them, the differential proportion vector corresponding to the laser ranging sequence is used as the elevation differential proportion vector; and the differential proportion vector corresponding to the first wave travel time sequence of the shallow road surface is used as the travel time differential proportion vector.

[0024] Preferably, in one feasible embodiment, the method for obtaining the pulse space offset matrix is ​​as follows: Considering that there is a rigid chassis physical distance between the front laser scanner and the rear ground-penetrating radar in actual practice, in order to mathematically restore the reference pulse span corresponding to this physical distance, the mechanical fixed distance is divided by the nominal driving mileage of a single pulse to obtain the theoretical delayed pulse number, and the standard pulse hysteresis difference under the ideal state without slippage is determined, which facilitates the subsequent benchmark anchoring of the deviation degree of the actually generated misaligned pulses; wherein, the nominal driving mileage of a single pulse is set according to the chassis hardware specifications and can be directly called; in order to accurately reflect the specific degree of deviation of the slippage redundant pulse from the chassis physical constraints, and further for any Given an original radar transmission sequence number and any reference sequence number, the absolute value of the difference between the original radar transmission sequence number and the reference sequence number, and then subtracting the theoretical delay pulse number, is used as the pulse space offset penalty value corresponding to the original radar transmission sequence number and the reference sequence number. This accurately reflects the objective algebraic cost of the radar's measured triggering timing violating the vehicle's rigid mechanical spacing. A larger pulse space offset penalty value indicates a more unreasonable forced matching between the original radar transmission sequence number and the reference sequence number. To form a global constraint model and establish the boundary conditions for optimal transmission, the reference sequence number is used as the row of a matrix, the original radar transmission sequence number as the column, and the pulse space offset penalty value as the element, generating a pulse space offset matrix, as shown below. Figure 2 The image shows the pulse space offset matrix. Since the total number of pulses generated by the ground-penetrating radar is always greater than the total number of laser reference pulses when the wheels slip, the pulse space offset matrix is ​​a rectangular matrix with fewer rows than columns.

[0025] Preferably, in one feasible implementation of this embodiment, the method for obtaining the pulse probability matrix is ​​as follows: Considering that the optimal transmission algorithm needs to transform the distance penalty algebra into similarity probability weights, in order to construct a smooth and bounded allocation optimization space, each element in the pulse space offset matrix is ​​subjected to negative exponential decay processing. Specifically, the mean of all elements in the pulse space offset matrix is ​​first obtained as a smoothing constant, and then... Each of the above elements is subjected to negative exponential decay, where M represents an element in the pulse space offset matrix. It is a smoothing constant; To pre-set a very small positive number (e.g., 0.001) to avoid the denominator being 0 when the smoothing constant is 0, the implementer can set it according to the actual situation, and there is no limitation here; exp is an exponential function with the natural constant as the base; thus generating a kernel matrix, which accurately reflects the matching feasibility kernel value between different pulse combinations (the larger the penalty, the closer the kernel value is to 0), which is beneficial to transform discrete rigid algebraic constraints into a continuous probabilistic solution base; To establish the initial computational state for the bidirectional scaling iteration, the left and right column vectors are initialized. This involves initializing all values ​​in the right column vector to 1 and creating an empty or arbitrary value cache for the left column vector to facilitate the initial alternating multiplication and division operations. Considering that the iterative solution requires fusing the spatial boundary penalty from the preceding steps with the current feature distribution on the radar side, to calculate the theoretical distribution expectation on the laser side in the current state, the kernel matrix is ​​multiplied by the right column vector. Each element of the multiplication result is added to a preset third positive number to generate the first denominator vector. In this embodiment, the preset third positive number is set to... To ensure that matrix division does not trigger the underlying division-to-zero overflow error, the implementer can set a preset third positive number according to the system's computing power environment, which is not limited here; in order to correct and scale the allocation expectation through the real feature distribution of the laser side, each element in the elevation difference proportion vector is divided by the corresponding element in the first denominator vector, and the left column vector is updated. Considering the mathematical structure balance of the bidirectional mapping, in order to similarly calculate the theoretical allocation expectation of the radar side under the current state, the transpose of the kernel matrix is ​​multiplied by the left column vector. Each element of the multiplication result is added to a preset fourth positive number to generate a second denominator vector. In this embodiment, the preset fourth positive number is set to... To ensure the denominator is strictly non-zero to guarantee the accuracy of reverse calculation, the implementer can set a preset fourth positive number according to the system's computing power environment, which is not limited here; in order to perform reverse correction scaling on the allocation expectation through the real feature distribution on the radar side, each element in the travel time difference proportion vector is divided by the element at the corresponding position in the second denominator vector, and the right column vector is updated. To gradually approach the optimal feature allocation path for heterogeneous multi-source data, the update operations of the left and right column vectors are alternately executed until a preset stopping condition is met, obtaining the converged left and right column vectors. In this embodiment, the preset stopping condition is set as follows: the sum of the absolute errors between the current left column vector and the left column vector from the previous iteration is less than a preset convergence constant, or the current update operation count reaches the maximum iteration count threshold. In this embodiment, the preset convergence constant is set as follows: To ensure a stable mapping state in terms of mathematical accuracy, a maximum iteration threshold of 1000 is set to prevent the algorithm from getting stuck in an infinite loop and exhausting the terminal's computing power under complex and extreme noise data. Implementers can set preset convergence constants and maximum iteration thresholds according to the performance of specific computing chips and the real-time requirements of detection; no restrictions are imposed here. To restore the one-dimensional feature correction vector to the global two-dimensional mapping framework, the converged left and right column vectors are successively expanded into a first diagonal matrix and a second diagonal matrix, i.e., the elements on the diagonal are the corresponding elements in the vector, and the remaining positions are 0; then, the first diagonal matrix, the kernel matrix, and the second diagonal matrix are multiplied sequentially to obtain the pulse probability matrix.

[0026] Step S3: Based on the target column index corresponding to the largest element in each row of the pulse probability matrix, extract the single-column waveform data corresponding to the original three-dimensional echo of the ground penetrating radar, and stitch them together to generate a spatiotemporal echo matrix; obtain the location of the defect and its local correction coefficient based on the laser ranging change combined with the spatiotemporal echo matrix; obtain the remaining safe distance based on the reflection travel time of the waveform at the defect location and the local correction coefficient.

[0027] Specifically, considering that the high-frequency superposition of ground-penetrating radar electromagnetic waveforms can easily lead to destructive interference, in order to preserve the original high-frequency signal used for defect detection and eliminate spatiotemporal misalignment, single-column waveform data corresponding to the original three-dimensional echo of the ground-penetrating radar is extracted based on the target column index corresponding to the largest element in each row of the pulse probability matrix. This data is then stitched together to generate a spatiotemporal echo matrix, accurately reflecting the spatial fusion correlation between the geometric undulations of the road surface and the internal electromagnetic reflection sequence under the same geographic coordinate reference. This achieves a one-to-one correspondence between the original data coordinates and constructs a coaxial data base. Considering that the road surface in areas prone to slippage is uneven and easily accompanied by water infiltration, in order to overcome the depth calculation error caused by fixed empirical wave velocity, the defect location and its local correction coefficient are obtained based on the laser ranging variation combined with the spatiotemporal echo matrix. This accurately locates the specific spatial characteristic coordinates of the area requiring maintenance intervention and reflects the local equivalent attenuation characteristics of the ground-penetrating radar wave transmission speed due to the loose material and moisture accumulation state inside the defect point. In order to accurately quantify the actual depth of the disease extending downwards to define the integrity of the bearing structure, and then obtain the remaining safe distance based on the reflection travel time of the waveform at the location of the disease and the local correction coefficient, the true physical thickness of the bottom of the disease from the interface of the underlying healthy structure is accurately reflected. This is beneficial to provide a reliable basis for the subsequent development of precise anti-recurrence cutting schemes and the generation of digital mechanical control commands.

[0028] Preferably, in one feasible method of this embodiment, the method for obtaining the spatiotemporal echo matrix is ​​as follows: For any row in the pulse probability matrix, the column index corresponding to the largest element in that row is taken as the target column index of that row, because the largest element in that row represents the highest probability and most reasonable mapping of the radar emitted pulse to the current laser reference number after comprehensively considering the distribution of abrupt change characteristics and physical spatial distance constraints; if there are at least two column indices corresponding to the largest element in that row (i.e., optimization deadlock caused by the homogenization of data in the smooth road section), then each column index is obtained by subtracting the corresponding reference number, and then subtracting the theoretical delay pulse number to obtain the target column index. The reference degree of the column index accurately reflects the specific algebraic deviation of the corresponding candidate radar pulses from the physical travel distance under normal, slip-free conditions in terms of time span. When the reference degree is closer to 0, it indicates that although the corresponding radar pulse is equivalent to other candidates in terms of feature similarity, its physical trigger position best matches the nominal delay state of the chassis rigid mechanical spacing, making the match the most reliable. Therefore, the column index corresponding to the minimum absolute value of the reference degree is taken as the target column index of that row. By introducing a deviation penalty minimization strategy, the multi-peak optimization deadlock is resolved. If there are at least two column indices corresponding to the minimum value, then any one of the column indices is selected as the target column index. To avoid the cancellation of peaks and troughs or high-frequency phase distortion caused by directly accumulating the chaotic and redundant slippage waveforms, this method extracts single-column waveform data corresponding to all target column indices from the original 3D echo of the ground-penetrating radar. This preserves the original amplitude shape of the waveforms. Then, all extracted single-column waveform data are sequentially concatenated according to their corresponding reference numbers to generate a spatiotemporal echo matrix. This completely transforms the original ground-penetrating radar detection sequence, which contained a large amount of redundancy and misalignment, into a data base perfectly coaxially aligned with the laser reference coordinates. To visually demonstrate the alignment effect of the above algorithm under slippage conditions, as shown... Figure 3 The image shown is a heatmap for solving the pulse probability matrix under wheel slippage conditions according to an embodiment of the present invention. Figure 3 When a vehicle drives through a pothole and waterlogged area, its chassis slips. The radar's original transmitted pulses locally reach 350, while the actual corresponding laser reference pulses are only 200. The dark, bright ridges in the figure represent the target column index mapping trajectory after alternating scaling and iterative convergence. The results show that by constraining the spatial offset cost, the nonlinear misalignment caused by 150 slippage redundant pulses is successfully converged, the divergent mapping is truncated, and the reconstruction accuracy of the spatiotemporal echo matrix is ​​guaranteed.

[0029] Preferably, in one feasible embodiment of this method, the location of the damage is obtained as follows: Considering that road surface potholes and damage manifest as significant spatial concave extreme values ​​in the laser ranging sequence, in order to pinpoint the feature coordinates requiring repair intervention, extreme ranging points are extracted from the laser ranging sequence. The geographic coordinates corresponding to the reference number of each extreme ranging point are all taken as the location of the damage, establishing a correspondence between the damage points and a globally unified geographic mileage. It should be noted that, considering the possibility of spilled materials or raised debris on the road surface causing interference, in order to ensure the accuracy of positioning, negative extreme points caused by a decrease in ranging distance are removed during extraction based on direction determination, and only positive extreme points caused by an increase in ranging distance are retained as extreme ranging points.

[0030] Preferably, in one feasible method of this embodiment, the method for obtaining the local correction coefficient is as follows: for any defect location, the average value of laser ranging within a preset adjacent range of the extreme ranging point corresponding to the defect location is used as the benchmark elevation corresponding to the defect location, because the average value of laser ranging within the preset adjacent range objectively represents the original healthy road surface level before the damage occurred; in this embodiment, the preset adjacent range is set to 10 consecutive sampling points before and after the extreme ranging point to ensure that the data used for calculation are all in a flat and healthy area without defects. The implementer can set the preset adjacent range according to the laser sampling frequency, which is not limited here; it should be noted that, considering the situation where the detection vehicle has just started and the previous data points have not accumulated to the full span of the preset adjacent range, in order to prevent index out-of-bounds collapse during data frame extraction, an edge value copying and filling mechanism is used to fill in the previous boundary sequence (that is, copy and fill the first valid laser ranging value to the missing index position) to ensure the continuity and stability of the benchmark average calculation. Considering that abrupt changes in road surface geometry are directly related to internal structural defects, in order to quantify the geometric dimensions of road surface spalling, the absolute value of the difference between the extreme distance point corresponding to the defect location and the benchmark elevation is used as the depression depth of the defect location, accurately reflecting the physical spatial vertical drop of the defect location. When the depression depth is less than or equal to the preset depth threshold, it indicates that there is no significant structural spalling or macroscopic crack opening at the defect location. At this time, the preset wave velocity constant is used as the local correction coefficient for the defect location. In this embodiment, the preset wave velocity constant is set to 100 mm / ns to ensure that the nominal electromagnetic wave conversion velocity of the dry medium of the non-defective conventional asphalt pavement is directly substituted at the defect location, preventing over-correction calculation caused by shallow scratches. Implementers can set the preset wave velocity constant according to the type of pavement material, which is not limited here. When the depth of the depression exceeds a preset depth threshold, it indicates substantial structural damage at the location of the defect, which is highly susceptible to moisture infiltration and loosening of the internal medium. To extract equivalent electromagnetic wave attenuation characteristics, a single-column waveform data corresponding to the defect location is extracted from the spatiotemporal echo matrix. The absolute value of the difference between the travel time of the first peak signal exceeding the preset background threshold in this single-column waveform data and the reference reflection travel time of the healthy road surface is used as the shallow defect travel time. This excludes the absolute flight time of pure air gaps, accurately obtaining the abnormal relative delay caused by internal loosening and voiding. The reference reflection travel time of the healthy road surface is the average travel time of the first radar wave in the undamaged area within a preset range of the defect location. In this embodiment, the preset background threshold is set to three times the root mean square value of the background noise in the air section to ensure effective filtering of electromagnetic crosstalk and environmental noise encountered by radar waves propagating in the air. To accurately capture the initial strong reflection signal caused by the real physical interface, the implementer can set a preset background threshold based on the radar antenna's receiving gain and the level of electromagnetic interference in the field, without limitation here. The preset range is set to 10 consecutive undamaged radar sampling columns before and after the defect location to ensure that the obtained reference travel time is completely within the healthy area without defects and with normal compaction, thus providing a clean comparison scale. The implementer can set the preset range based on the trigger sampling interval of the radar along the survey line, without limitation here. It should be noted that, considering that the defect location is at the beginning and end of the road segment, which may cause the radar sampling columns before and after to not accumulate to the full span of the preset range, in order to prevent index out-of-bounds during data extraction, an edge value copying and filling mechanism is used to fill in the boundary sequence to ensure the stability of the reference reflection travel time. Considering the potential for discrepancies in physical properties when substituting cross-medium wave velocity, leading to distortion in the calculation of the remaining safety distance, this embodiment aims to accurately derive the wave velocity attenuation characteristics within the solid medium. The sum of the healthy pavement reference reflection travel time and the shallow damage travel time is used as the first result. Dividing the healthy pavement reference reflection travel time by the first result yields the wave velocity attenuation factor, accurately reflecting the relative reduction in electromagnetic wave propagation efficiency due to internal medium degradation and moisture accumulation. To obtain the equivalent electromagnetic wave velocity reflecting the true deterioration state of the local solid medium, a preset wave velocity constant is multiplied by the wave velocity attenuation factor to obtain the local correction coefficient for the location of the defect. This completely eliminates the depression depth difference variable representing the external air and water environment. By scaling the nominal healthy wave velocity of the defect-free section using the waveform delay time ratio, the physical parameters within the homogeneous structure are corrected, ensuring that subsequent thickness calculations are not distorted. This embodiment sets a preset depth threshold of 10mm to effectively filter out minor undulations caused by the roughness of the conventional pavement texture. Implementers can set the preset depth threshold according to the pavement surface layer thickness; this is not limited here.

[0031] Preferably, in one feasible way of this embodiment, the method for obtaining the remaining safe distance is as follows: considering that the stability of the underlying load-bearing structure is the core basis for evaluating the milling scheme, in order to define the limit depth of the disease invading the current structural layer, for any disease location, extract the single-column waveform data corresponding to the disease location in the spatiotemporal echo matrix, use the amplitude extreme value change characteristics of the reflection envelope, set a time domain shielding window starting from the first wave travel time of the shallow road surface at the disease location plus the preset blind zone delay, within the waveform segment range after avoiding the time domain shielding window, obtain the time corresponding to the first feature point exceeding the warning amplitude threshold from the single-column waveform data, as the first reflection travel time corresponding to the bottom of the disease, and within the set adjacent healthy structural layer depth judgment window, obtain the second reflection travel time corresponding to the interface of the adjacent healthy structural layer based on the zero-crossing jump point of the first derivative, and lock the echo position of the stable boundary of the bottom surface of the structure and the tip of the disease; The purpose of the preset blind zone delay is to ensure that the search starting point is free from the interference of the strongest initial reflection segment at the natural air interface. In this embodiment, it is set to 2ns to ensure that the algorithm can completely skip the pulse oscillation tail area of ​​the main reflection wave from the road surface, avoiding misjudging the echo of shallow peeling as the bottom of deep-seated defects. The implementer can set the preset blind zone delay according to the center frequency and transmitted pulse width of the ground penetrating radar antenna, which is not limited here. In this embodiment, the warning amplitude threshold is set to 3 times the root mean square value of the background noise of the single-column waveform data to ensure effective filtering of electromagnetic interference caused by small clutter at the bottom layer and the inhomogeneity of conventional materials, thereby accurately capturing the defects caused by the disease. The actual structural reflection signal caused by the sudden change in the dielectric constant at the bottom can be set by the implementer based on the receiving gain of the ground penetrating radar and the average electromagnetic attenuation characteristics of the road surface material. No limit is set here. The depth judgment window of the adjacent healthy structural layer is set to the ±20% time domain interval of the nominal structural layer thickness corresponding to the travel time in the historical pavement record of the target road. This ensures that the peak-finding algorithm only performs effective retrieval near the target load-bearing interface and avoids erroneously capturing the reflection signal of deeper subgrade or shallow unrelated interface. The implementer can set the depth judgment window of the adjacent healthy structural layer based on the as-built design documents of the specific road and the radar sampling frequency. No limit is set here. To obtain the pure internal relative delay travel time of the structure, the second reflection travel time is subtracted from the first reflection travel time to obtain the time difference value at the location of the defect, eliminating the fixed travel time interference from the upper structure. Considering that the local correction coefficient has already equivalently included the characteristic variables of water-rich and loose states, in order to convert the time dimension data into the true physical thickness, the product of the time difference value at the location of the defect and the local correction coefficient is divided by a constant 2 to obtain the remaining safe distance at the location of the defect. It should be noted that the division by a constant 2 is because the first and second reflection travel times obtained by ground penetrating radar are both the two-way propagation time of electromagnetic waves from the transmitting antenna downwards to the reflecting surface and back to the receiving antenna, while the actual physical distance is only the one-way depth. Therefore, the product must be halved to restore the true remaining cutting thickness. To verify the accuracy of the above local correction coefficient compared with the traditional fixed empirical wave velocity in depth calculation, such as... Figure 4 The image shown is a comparison chart of milling depth calculation errors in a water-rich crack region according to an embodiment of the present invention. Figure 4 In the pilot test, the average depth of the road surface depression was set at 15mm, and loose water was present inside. Traditional methods use fixed empirical wave velocities, which cannot reflect the additional delay of electromagnetic waves caused by moisture, resulting in a serious underestimation of the remaining safety distance, with a root mean square error as high as 12.4mm. However, after using the local correction coefficients dynamically calculated in this embodiment for travel time conversion, the reduction in electromagnetic wave velocity is physically equivalently compensated, and the calculation error of the remaining safety distance is significantly reduced to 2.1mm. The above data proves that the remaining safety distance calculated in this embodiment is more accurate, providing a guarantee for the subsequent generation of reliable cross-layer cutting depth parameters.

[0032] Step S4: Generate milling instructions based on the remaining safety distance and geographic coordinates.

[0033] Specifically, the known remaining safety distance objectively reflects the physical thickness from the bottom of the defect to the interface of the next healthy structural layer. In order to issue clear cutting action parameters to the on-site automated pavement repair machine, and then generate milling instructions based on the remaining safety distance and geographical coordinates, the entire process of distance alignment, depth assessment and milling decision is automated and closed-loop control is realized, effectively eliminating the hidden dangers of underlying defects.

[0034] Preferably, in one feasible method of this embodiment, the method for generating milling instructions is as follows: For any defect location, when the remaining safe distance of the defect location is greater than or equal to the preset cutting buffer threshold, it indicates that the defect distribution is relatively shallow, the bottom of the crack is still far from the underlying load-bearing structure with sufficient safe physical thickness, and the defect has not pierced the current asphalt surface layer. In order to avoid excessive milling and waste of healthy pavement material and to ensure the integrity of the underlying load-bearing capacity, the preset single-layer cutting depth parameter is retrieved, instructing the on-site milling equipment to remove only the top layer of pavement structure. In order to achieve accurate anchoring of equipment control parameters and actual physical working space, the preset single-layer cutting depth parameter is then spliced ​​with the geographic coordinates corresponding to the defect location to generate milling instructions. In this embodiment, the preset single-layer cutting depth parameter is set to a fixed thickness value of 40mm to ensure that the cutting instruction can completely peel off the top fine-grained asphalt pavement wearing layer without damaging the intermediate layer. The implementer can set the preset single-layer cutting depth parameter according to the standard design thickness of the upper layer in the historical pavement record of the target road, which is not limited here. When the remaining safe distance at the location of the defect is less than the preset cutting buffer threshold, it indicates that the cracks and other defects have approached or even completely penetrated the current structural layer, and the stress concentration at the bottom is severe. Single-layer milling can no longer effectively prevent the further downward extension of microcracks. In order to thoroughly eliminate the root cause of the defect and prevent the defect from recurring rapidly after the road is opened to traffic, the preset standard milling depth is added to the preset structural layer thickness to generate a cross-layer cutting depth parameter to increase the physical cutting range of the facade. Then, the cross-layer cutting depth parameter is spliced ​​with the geographical coordinates corresponding to the location of the defect to generate a milling command. This embodiment sets a fixed thickness value of 40mm for the preset standard milling depth to ensure that this serves as the reference cutting starting point for the current damaged structural layer. The implementer can set the preset standard milling depth according to the actual completed thickness of the wear layer on the target road section surface, which is not limited here. The preset structural layer thickness is also set to a fixed thickness value of 40mm. The implementer can set the preset structural layer thickness according to historical paving records, which is not limited here. The preset cutting buffer threshold is set to 40mm to ensure that it matches the standard paving thickness baseline of a single layer of asphalt surface in the current municipal road maintenance specifications, thereby reserving sufficient judgment buffer for the underlying structure. The implementer can set the preset cutting buffer threshold according to the specific road design level and historical paving records, which is not limited here.

[0035] In summary, this embodiment obtains the reference sequence numbers of laser ranging, the first wave travel time of the shallow road surface, and laser ranging, and binds them to geographic coordinates and the original radar transmission sequence number; it obtains the elevation and travel time difference ratio vector based on the changes in laser ranging and the first wave travel time of the shallow road surface; it constructs a pulse spatial offset matrix based on the fixed spacing and sequence number of the sensors, maps it to a kernel matrix, and obtains a pulse probability matrix by alternating scaling and iteration with the two ratio vectors, thereby extracting the original ground-penetrating radar waveform data and splicing it into a spatiotemporal echo matrix; based on the changes in laser ranging and the spatiotemporal echo matrix, combined with the reflection travel time, it obtains the location of the defect and its remaining safe distance, thereby generating milling commands. This invention effectively overcomes the data misalignment caused by wheel slippage and the error of fixed empirical wave velocity, providing a precise cutting command control closed loop for automatic road repair.

[0036] Example 2: This invention also proposes a non-destructive testing device for deep cracks in pavement repair. The device includes a memory and a processor. The memory stores executable program code, and the processor calls and executes this executable program code to perform a non-destructive testing method for deep cracks in pavement repair according to this invention. Specifically, the device may be a chip, component, or module. The chip may include a connected processor and memory; the memory stores instructions, and when the processor calls and executes the instructions, the chip can perform the non-destructive testing method for deep cracks in pavement repair provided in the above embodiments.

Claims

1. A non-destructive testing method for deep cracks in road surface repair, characterized in that, The method includes the following steps: Obtain the first wave travel time of the shallow surface layer of laser ranging and ground penetrating radar within a single road segment; obtain the reference sequence number of laser ranging and bind it to geographic coordinates; obtain the original radar transmission sequence number; Based on the changes in the first wave travel time of laser ranging and shallow road surface, the elevation difference proportion vector and travel time difference proportion vector within a single road segment are obtained sequentially; based on the mechanical fixed distance between the laser scanner and the ground penetrating radar in the longitudinal direction, the reference number, and the original radar transmission number, a pulse spatial offset matrix is ​​constructed; the pulse spatial offset matrix is ​​mapped to a kernel matrix, and alternately scaled and iterated with the elevation difference proportion vector and the travel time difference proportion vector to obtain the pulse probability matrix; Based on the target column index corresponding to the largest element in each row of the pulse probability matrix, single-column waveform data corresponding to the original three-dimensional echo of the ground penetrating radar are extracted and spliced ​​to generate a spatiotemporal echo matrix; the location of the defect and its local correction coefficient are obtained based on the laser ranging variation combined with the spatiotemporal echo matrix; the remaining safe distance is obtained based on the reflection travel time of the waveform at the defect location and the local correction coefficient. Based on the remaining safety distance and geographic coordinates, generate milling instructions.

2. The non-destructive testing method for deep cracks in road surface repair as described in claim 1, characterized in that, The methods for obtaining the elevation difference proportion vector and the travel time difference proportion vector are as follows: The laser ranging sequences are arranged in the order of acquisition to obtain the laser ranging sequence; the first wave travel times of the shallow road surface are arranged in the order of acquisition to obtain the first wave travel times of the shallow road surface. For any sequence in the laser ranging sequence and the shallow road surface first wave travel time sequence, obtain the first-order absolute difference of adjacent data in the sequence, and use the sum of all first-order absolute differences as the total difference value. The sum of each first-order absolute difference and a preset first positive number is used as the numerator, and the sum of the total difference and a preset second positive number is used as the denominator. Each numerator is divided by the denominator to obtain the difference proportion value and concatenate them into a difference proportion vector. The differential proportion vector corresponding to the laser ranging sequence is used as the elevation differential proportion vector; the differential proportion vector corresponding to the first wave travel time sequence of the shallow road surface is used as the travel time differential proportion vector.

3. The method for non-destructive testing of deep cracks in road surface repair as described in claim 1, characterized in that, The method for obtaining the pulse space offset matrix is ​​as follows: Divide the mechanical fixed interval by the nominal mileage of a single pulse to obtain the theoretical number of delayed pulses; For any original radar transmission sequence number and any reference sequence number, the absolute value of the difference between the original radar transmission sequence number and the reference sequence number, and then the theoretical delay pulse number, is used as the pulse space offset penalty value corresponding to the original radar transmission sequence number and the reference sequence number. The pulse space offset matrix is ​​generated by using the reference sequence number as the row of the matrix, the original radar transmission sequence number as the column of the matrix, and the pulse space offset penalty value as the element of the matrix.

4. The non-destructive testing method for deep cracks in road surface repair as described in claim 1, characterized in that, The method for obtaining the pulse probability matrix is ​​as follows: Each element in the pulse space offset matrix is ​​subjected to negative exponential decay to generate a kernel matrix; Initialize the left and right column vectors, multiply the kernel matrix with the right column vector, add each element of the multiplication result to the preset third positive number to generate the first denominator vector, divide each element in the elevation difference proportion vector by the element at the corresponding position in the first denominator vector, and update the left column vector. Multiply the transpose of the kernel matrix with the left column vector, add each element of the multiplication result to the preset fourth positive number to generate the second denominator vector, divide each element in the travel time difference percentage vector by the element at the corresponding position in the second denominator vector, and update the right column vector. Alternately perform update operations on the left and right column vectors until a preset stopping condition is met, and obtain the converged left and right column vectors; Expand the converged left and right column vectors into first and second diagonal matrices, respectively. Multiply the first diagonal matrix, the kernel matrix, and the second diagonal matrix in sequence to obtain the impulse probability matrix.

5. The non-destructive testing method for deep cracks in road surface repair as described in claim 1, characterized in that, The method for obtaining the spatiotemporal echo matrix is ​​as follows: For any row in the pulse probability matrix, the column index corresponding to the largest element in that row is taken as the target column index of that row; if there are at least two column indices corresponding to the largest element in that row, then each column index is subtracted from the corresponding baseline number, and then the theoretical delay pulse number is subtracted to obtain the reference level of the corresponding column index; the column index corresponding to the minimum absolute value of the reference level is taken as the target column index of that row. Extract single-column waveform data corresponding to all target column indices from the raw three-dimensional echo of the ground penetrating radar; All extracted single-column waveform data are concatenated sequentially according to their corresponding reference numbers to generate a spatiotemporal echo matrix.

6. The non-destructive testing method for deep cracks in road surface repair as described in claim 2, characterized in that, The method for obtaining the location of the disease is as follows: Extreme distance points are extracted from the laser ranging sequence, and the geographic coordinates corresponding to the reference number of each extreme distance point are taken as the location of the disease.

7. The non-destructive testing method for deep cracks in road surface repair as described in claim 6, characterized in that, The method for obtaining the local correction coefficient is as follows: For any location of disease, the average value of laser ranging within a preset adjacent range of the extreme ranging point corresponding to the location of disease is used as the reference elevation corresponding to the location of disease. The absolute value of the difference between the extreme distance measurement point corresponding to the location of the disease and the benchmark elevation is taken as the concavity depth of the location of the disease. When the depression depth is less than or equal to a preset depth threshold, the preset wave velocity constant is used as the local correction coefficient for the location of the defect. When the depression depth is greater than a preset depth threshold, extract the single-column waveform data corresponding to the location of the defect in the spatiotemporal echo matrix, and take the absolute value of the difference between the travel time of the first peak signal exceeding the preset background threshold in the single-column waveform data and the reference reflection travel time of the healthy road surface as the shallow damage travel time; wherein, the reference reflection travel time of the healthy road surface is the average value of the first wave travel time of the radar in the undamaged area within a preset range of the location of the defect; The sum of the reference reflection travel time of the healthy pavement and the travel time of the shallow damage is taken as the first result; the reference reflection travel time of the healthy pavement is divided by the first result to obtain the wave velocity attenuation factor; the preset wave velocity constant is multiplied by the wave velocity attenuation factor to obtain the local correction coefficient for the location of the defect.

8. The non-destructive testing method for deep cracks in road surface repair as described in claim 1, characterized in that, The method for obtaining the remaining safe distance is as follows: For any location of a defect, extract the single-column waveform data corresponding to that location from the spatiotemporal echo matrix. Utilize the extreme value mutation characteristics of the reflection envelope amplitude, set a time-domain shielding window starting from the first wave travel time of the shallow road surface layer at that defect location plus a preset blind zone delay. Within the waveform segment range after avoiding the time-domain shielding window, obtain the time corresponding to the first feature point exceeding the warning amplitude threshold from the single-column waveform data as the first reflection travel time corresponding to the bottom of the defect. Within the set adjacent healthy structure layer depth judgment window, obtain the second reflection travel time corresponding to the interface of the adjacent healthy structure layer based on the zero-crossing jump point of the first derivative. Subtract the first reflection time from the second reflection time to obtain the time difference value of the disease location; The remaining safe distance at the location of the defect is obtained by multiplying the time difference at the defect location by the local correction coefficient and dividing by a constant 2.

9. The non-destructive testing method for deep cracks in road surface repair as described in claim 1, characterized in that, The method for generating milling instructions based on the remaining safety distance and geographic coordinates is as follows: For any defect location, when the remaining safe distance of the defect location is greater than or equal to the preset cutting buffer threshold, the preset single-layer cutting depth parameter is spliced ​​with the geographic coordinates corresponding to the defect location to generate a milling command. When the remaining safe distance at the location of the defect is less than the preset cutting buffer threshold, the preset standard milling depth is added to the preset structural layer thickness to generate the cross-layer cutting depth parameter. The cross-layer cutting depth parameter is concatenated with the geographic coordinates corresponding to the location of the defect to generate a milling command.

10. A non-destructive testing device for deep cracks in road surface repair, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the non-destructive testing method for deep cracks in road surface repair as described in any one of claims 1-9.

Images