Concrete internal defect sample generation method and system based on electromagnetic forward simulation

Abstract

The application discloses a concrete internal defect sample generation method and system based on electromagnetic forward simulation, and relates to the technical field of concrete structure nondestructive testing. The method comprises the following steps: an electromagnetic simulation area of a concrete structure to be detected is constructed, and electromagnetic simulation configuration parameters are obtained through constraint sampling, preset assignment and correlation calculation; electromagnetic forward simulation and solving are performed on a parameterized electromagnetic simulation model to generate A-scan and B-scan data; physical consistency labeling information is generated according to the relationship among physical true values, medium electromagnetic parameters, antenna positions, time zero points and electromagnetic wave propagation time delays, and the physical consistency labeling information is subjected to physical consistency quality constraint screening to form a labeling sample set. The application can batch generate concrete structure ground penetrating radar simulation labeling samples with physical interpretability, parameter traceability and label consistency.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for concrete structures, and in particular to a method and system for generating internal defect samples of concrete based on electromagnetic forward modeling. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Currently, internal defects in concrete structures are difficult to identify through external images. Ground-penetrating radar (GPR) has become an important means of identifying internal defects in concrete structures because it can perform non-destructive testing by utilizing reflected echoes generated by differences in the electromagnetic parameters of the medium. However, obtaining radar data with accurate physical values ​​in real-world engineering projects is extremely difficult: the true parameters of internal defects require destructive verification such as core drilling and excavation, which is costly and has limited sample size; at the same time, B-scan images are affected by hyperbolic superposition, strong clutter, and multipath interference caused by targets such as reinforcing bars and voids, and manual annotation is highly subjective, easily leading to misjudgments and inconsistencies.

[0004] In addition, existing simulation datasets are mostly geared towards roads or single scenarios, typically containing only simple targets or regular reinforcing bars, lacking comprehensive simulation of complex engineering constraints such as local curved surfaces, protective layer thickness, prestressed tendon arrangement, construction offset, corrosion layer, irregular voids, and aggregate clutter. Furthermore, most simulations only output radar images, lacking synchronous automatic conversion from physical ground truth to A-scan time windows, B-scan detection boxes, segmentation masks, and metadata labels, making it difficult to support the training needs of reproducible experiments and deep learning models. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for generating concrete internal defect samples based on electromagnetic forward modeling. This method can generate in batches ground-penetrating radar simulation labeled samples of concrete structures with physical interpretability, parameter traceability, and label consistency. These samples can be used for defect identification model training, pre-training, domain adaptation, and equipment algorithm verification in engineering scenarios that require the detection of internal defects in concrete.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: The first aspect of this invention provides a method for generating concrete internal defect samples based on electromagnetic forward modeling, comprising the following steps: An electromagnetic simulation region for the concrete structure to be tested is constructed, and electromagnetic simulation configuration parameters are obtained through constraint sampling, preset assignment, and correlation calculation. Based on the defect-related parameters, a parameterized electromagnetic simulation model is constructed, and electromagnetic forward modeling and solution are performed on the parameterized electromagnetic simulation model to generate A-scan and B-scan data. Based on the physical truth value, medium electromagnetic parameters, antenna position, zero point of time, and electromagnetic wave propagation delay relationship, physical consistency labeling information corresponding to the A-scan arrival time window, B-scan echo response region, and optional segmentation region is generated. The physical consistency labeling information is then filtered by physical consistency quality constraints to form a labeling sample set.

[0007] Furthermore, the specific steps for constructing the electromagnetic simulation region of the concrete structure to be tested are as follows: Establish a coordinate system for the local detection area of ​​the concrete structure to be inspected, and uniformly represent the detection window of the plane, curved or irregular surface as an electromagnetic simulation area in the scanning distance direction, depth direction and thickness or lateral direction.

[0008] Furthermore, the engineering structural constraints include at least the component thickness range, the protective layer thickness range, the set of spacing of the reinforcing mesh or prestressing tendons, the set of diameters of the reinforcing bars or prestressing tendons, the overall initial offset of the reinforcing bars, the single-strand binding or construction offset, the survey line range, and optional parameters of construction joints, delamination interfaces, or defect prior areas; the material electromagnetic constraints include at least the sampling range of the relative permittivity and conductivity of concrete, the ideal conductor or highly conductive material parameters of the reinforcing bars, the dielectric parameters of air, cracks, or cavities, and the equivalent dielectric parameters and conductivity of the corrosion layer.

[0009] Furthermore, a finite-difference time-domain electromagnetic forward modeler is used to solve the parameterized electromagnetic simulation model. The finite-difference time-domain electromagnetic forward modeler solves the Maxwell equations for lossy media on the Yee grid and satisfies the spatial step size, time step size, and Courant stability conditions.

[0010] Furthermore, the specific steps for performing electromagnetic forward modeling and solving on the parametric electromagnetic simulation model to generate A-scan and B-scan data are as follows: The transmitting and receiving antennas are moved sequentially along the preset measurement line. The finite-difference time-domain electromagnetic forward modeling solver is invoked to output the time series of the received electric field for each measurement point as an A-scan. Multiple A-scans on the same measurement line are stacked into a B-scan matrix according to the measurement point order. The B-scan matrix is ​​then subjected to device domain degradation, amplitude calibration, background removal, time-varying gain, normalization, and image resampling.

[0011] Furthermore, the physical consistency quality constraint screening includes checking whether the target is within the coverage area of ​​the survey line, whether the target echo time is within the sampling window, whether the B-scan energy is abnormal, whether the target echo peak is visible, whether the signal-to-noise ratio is lower than the threshold, whether the echo response outer region exceeds the imaging field of view, and whether the sample ratio of rebar, voids, cracks, and rust layers meets the preset dataset distribution.

[0012] A second aspect of the present invention provides a concrete internal defect sample generation system based on electromagnetic forward modeling simulation, comprising: The parameter sampling module is used to construct the electromagnetic simulation area of ​​the concrete structure to be tested, and obtain the electromagnetic simulation configuration parameters through constraint sampling, preset assignment and correlation calculation. The forward simulation module is used to construct a parameterized electromagnetic simulation model based on defect-related parameters, and to perform electromagnetic forward simulation and solve the parameterized electromagnetic simulation model to generate A-scan and B-scan data. The physical label generation and filtering module is used to generate physical consistency labeling information corresponding to the A-scan arrival time window, B-scan echo response area, and optional segmentation area based on the physical truth value, medium electromagnetic parameters, antenna position, time zero point, and electromagnetic wave propagation time delay relationship. It also performs physical consistency quality constraint filtering on the physical consistency labeling information to form a labeling sample set.

[0013] A third aspect of the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and executed the steps of the method for generating concrete internal defect samples based on electromagnetic forward modeling as described in the first aspect of the present invention.

[0014] A fourth aspect of the present invention provides a computer device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the method for generating concrete internal defect samples based on electromagnetic forward modeling as described in the first aspect of the present invention.

[0015] A fifth aspect of the present invention provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the method for generating concrete internal defect samples based on electromagnetic forward modeling as described in the first aspect of the present invention.

[0016] The above one or more technical solutions have the following beneficial effects: This invention discloses a method and system for generating concrete internal defect samples based on electromagnetic forward modeling. It explicitly incorporates engineering physical constraints such as component thickness, protective layer thickness, reinforcement or prestressing tendon arrangement, construction offset, material dielectric parameters, and defect morphology into the scene generation process. Compared to simulation methods that simply generate geometry randomly, this method can obtain radar echoes that more closely resemble actual concrete structures. This invention binds physical truth values ​​to simulated geometric objects and automatically generates A-scan and B-scan labels using electromagnetic wave propagation delay formulas, reducing manual annotation errors and ensuring consistency of the same defect across physical space, time axis, and image axis. This invention can batch generate ground-penetrating radar simulation annotation samples of concrete structures with physical interpretability, parameter traceability, and label consistency. These samples can be used for defect identification model training, pre-training, domain adaptation, and equipment algorithm verification in engineering scenarios requiring the detection of internal concrete defects.

[0017] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the method for generating concrete internal defect samples based on electromagnetic forward modeling in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of the local planar / curved surface development coordinates and local FDTD model of the concrete structure in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the physical constraint modeling of targets and defects inside a concrete structure in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the electromagnetic forward modeling, device domain degradation, and image enhancement link in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram illustrating the mapping from physical truth values ​​to A-scan / B-scan tags in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the simulation dataset generation and data output process in Embodiment 1 of the present invention; Figure 7This is a schematic diagram of the simulation sample, B-scan, and detection tag in Embodiment 1 of the present invention; Figure 8 This is a schematic diagram of the simulation geometric model in Embodiment 1 of the present invention: reinforcing bars, voids, cracks, and aggregate clutter; Figure 9 This is a B-scan radar image generated from the simulation of the geometric model in Embodiment 1 of the present invention; Figure 10 This is a schematic diagram of the automatic annotation of the simulated B-scan radar image in Embodiment 1 of the present invention. Detailed Implementation

[0020] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] Example 1: Embodiment 1 of the present invention provides a method for generating concrete internal defect samples based on electromagnetic forward modeling simulation, such as... Figure 1 and Figure 6As shown, this invention transforms the local planar, curved, or irregular detection window of the concrete structure to be inspected into a two-dimensional or quasi-three-dimensional simulation window. Within this window, based on physical parameters, it samples and generates the concrete matrix, reinforcing bars or prestressed tendons, corrosion extension layers, irregular voids, broken-line cracks, void interfaces, honeycomb regions, and aggregate clutter. Each geometric object synchronously records its physical center, boundary range, category, material number, and electromagnetic parameters during generation, forming a traceable truth value from the source. Then, the parameterized scene is converted into an FDTD solver input, and the transmitting / receiving antenna is moved along the survey line to solve the received electric field components point by point, forming an A-scan sequence and a B-scan matrix. Furthermore, the forward modeling matrix is ​​converted into an image consistent with the statistical characteristics of actual radar equipment through noise superposition, bandpass filtering, time zero-point blanking, amplitude calibration, background removal, and time-varying gain.

[0023] Finally, the propagation speed is calculated based on the dielectric constant of concrete, and the echo arrival time is derived based on the air layer path, defect depth, and zero-point time delay. The physical space coordinates are automatically mapped to the A-scan time window and the B-scan normalized target box, and the original matrix, enhanced image, label file, and metadata are saved uniformly.

[0024] Specifically, the following steps are included: Step 1: Construct the electromagnetic simulation area of ​​the concrete structure to be tested, and obtain the electromagnetic simulation configuration parameters through constraint sampling, preset assignment and correlation calculation.

[0025] Step 1.1: Construct the electromagnetic simulation area of ​​the concrete structure to be tested.

[0026] In one specific implementation, a coordinate system is established for the local detection area of ​​the concrete structure to be inspected, and the detection windows of planar, curved, or irregular surfaces are uniformly represented as electromagnetic simulation areas in the scanning distance direction, depth direction, and thickness or lateral direction.

[0027] Specifically, such as Figure 2 As shown, the coordinate system of the local detection region satisfies the following for the curved surface structure. For planar or straight survey line structures, the following conditions must be met. ,in Let be the local radius of curvature of the component. For the surface parameter angular coordinates, For the distance of the survey line, The depth coordinates are the points from the surface of the component to the interior of the component. The thickness direction or the direction that remains unchanged in 2D / quasi-3D simulation is used; and the surface type, radius of curvature, unfolded window length and local planarization error are recorded in the metadata; when the object to be detected is a planar component, the radius of curvature is set to infinity or the planarization error is recorded as 0.

[0028] For the detection window length L is much smaller than the radius of curvature In such scenarios, a local planarization approximation can be used. After mapping the arc length of the curved surface to a plane x-coordinate, the local geometric error can be estimated by the chord-arc difference: (1).

[0029] when When the mesh step size is smaller than the annotation tolerance, a two-dimensional or quasi-three-dimensional model can be used to improve the efficiency of batch simulation; when When the value is large, record it in the metadata. The simulation domain includes an air layer at the top and a concrete layer at the bottom, with the boundary set as a perfectly matched layer or other absorbing boundary. It also includes the calculation of expansion errors and the addition of curvature correction in the coordinate mapping.

[0030] Step 1.2: Obtain electromagnetic simulation configuration parameters through constraint sampling, preset assignment, and correlation calculation.

[0031] In one specific implementation, structural constraints include at least the component thickness range, protective layer thickness range, set of spacing for reinforcing mesh or prestressed tendons, set of diameters for reinforcing bars or prestressed tendons, overall initial offset of reinforcing bars, single-strand binding or construction offset, survey line range, and optional parameters for construction joints, delamination interfaces, or priori defect regions. Electromagnetic constraints include at least the sampling range for the relative permittivity and conductivity of concrete, parameters for ideal conductors or highly conductive materials in reinforcing bars, dielectric parameters for air, cracks, or cavities, and equivalent dielectric parameters and conductivity of the corrosion layer. Defect geometric constraints limit the type, quantity, location, size, orientation, and morphology of defects in physical space. Equipment acquisition constraints mean that the antenna, survey line, sampling, and imaging range during simulation must be consistent with the predetermined ground-penetrating radar acquisition method.

[0032] Electromagnetic simulation configuration parameters include dielectric electromagnetic parameters, component and defect geometric parameters, heterogeneous background parameters, antenna parameters, and scanning acquisition parameters.

[0033] Specifically, the following parameters are included: Concrete parameters: These are dielectric electromagnetic parameters, including relative permittivity, conductivity, magnetic permeability, and the range of the concrete region; the permittivity is 6.0 to 8.5, and the conductivity is 0.005 to 0.012 S / m.

[0034] Reinforcing bar parameters: These belong to the component's geometric parameters and material electromagnetic parameters, including the reinforcing bar position, protective layer thickness, spacing, diameter, initial offset, binding offset, and metal material number.

[0035] Corrosion layer parameters: These belong to the defect geometry parameters and material electromagnetic parameters, including the location of corrosion occurrence, the outward expansion radius or expansion ratio, boundary perturbation, equivalent dielectric constant, and conductivity. The outward expansion radius is 1.8 to 2.5 times the radius of the reinforcing bar.

[0036] Void parameters: These belong to the geometric parameters of defects and the electromagnetic parameters of materials, including quantity, center coordinates, size, number of subclusters, number of boundary points, boundary disturbance, circumscribed dimensions, and void medium parameters.

[0037] Crack parameters: These belong to the geometric parameters of defects and the electromagnetic parameters of materials, including quantity, center coordinates, length, orientation angle, equivalent thickness, path disturbance, circumscribed dimensions, and crack medium parameters.

[0038] Aggregate clutter parameters: These are spatial distribution parameters of the heterogeneous medium inside concrete, including random field, smoothing scale, threshold, aggregate material number, and electromagnetic parameters; they are generated by thresholding after Gaussian smoothing of the random field.

[0039] Antenna parameters include: transmit waveform, center frequency, transmit and receive antenna positions, antenna spacing, antenna height above the surface, and received electric field components.

[0040] Scanning acquisition parameters include the start and end positions of the measurement line, measurement point step size, number of measurement points, time sampling interval, number of sampling points, time window, maximum detection depth, and time zero-point delay; the default number of measurement points is 200, the target number of sampling points is 1697, and the measurement point step size is 0.004 m.

[0041] Among them, aggregate clutter parameters are generated through smoothed random fields, random medium particles, or fractal medium fields; the corrosion layer is established in the form of an outer cladding layer around the steel bar, and the equivalent radius of the corrosion layer parameters is a preset multiple of the steel bar radius or a corrosion rate function; internal cavities are generated by superimposing multiple circular, elliptical, polygonal, or voxel clusters to form irregular boundaries; internal cracks are generated by broken line segments, continuous micro cylinders, thin voxels, or linear cavities, and the defect center, size, inclination angle, control point, circumscribed boundary, category number, and electromagnetic parameters are recorded.

[0042] The core of this embodiment lies in not randomly generating defects in isolation, but rather constraining sampling based on the engineering structure of the concrete structure to be tested, the electromagnetic properties of the materials, and the equipment acquisition conditions. In one embodiment, the relative permittivity of concrete is... Sampling within the range of 6.0 to 8.5, conductivity Sampling should be performed within the range of 0.005 S / m to 0.012 S / m; the concrete cover thickness should be sampled within the range of 50 mm to 70 mm; the rebar spacing should be selected from 150 mm and 200 mm; the rebar radius should be selected from 10 mm and 12.5 mm; and millimeter-level tying offsets should be added to each rebar. The above ranges are for illustrative purposes only and can be reconfigured according to the actual concrete component design drawings, testing specifications, and measuring equipment.

[0043] The physical location of the reinforcing bars can be represented as: (2).

[0044] in, The physical location of the reinforcing bars. This is the starting position of the reinforcing mesh. For the spacing of the reinforcing bars, To detect the surface coordinates of the component, For the thickness of the protective layer, and For binding offset. The corrosion layer can be defined as the area surrounding the reinforcing bar, with an equivalent radius of: (3).

[0045] in, Let the outer radius of the equivalent corrosion coating around the p-th rebar be . Let be the radius of the p-th rebar, where p is the rebar number. This represents the corrosion expansion magnification. Cavities are formed by the superposition of multiple random polygons, ellipses, or voxel clusters; cracks are represented by broken line segments or continuous thin-sheet voxels. Aggregate clutter can be generated by a smoothed random field. (4).

[0046] in, This is a spatial indicator field for aggregate clutter or heterogeneous media within concrete. 1 indicates that the location is classified as an aggregate or heterogeneous region, and 0 indicates ordinary concrete background. The code generates this field by Gaussian smoothing using a random field followed by thresholding. This indicates a Gaussian smoothing operation or a Gaussian filter. For a uniform random field, Gaussian smoothing scale, The threshold is set. A non-uniform concrete background is created using the above method, causing the B-scan to no longer exhibit an ideally uniform medium response.

[0047] Step 2: Construct a parameterized electromagnetic simulation model based on the defect-related parameters, and perform electromagnetic forward modeling and solution on the parameterized electromagnetic simulation model to generate A-scan and B-scan data.

[0048] Step 2.1: Construct a parameterized electromagnetic simulation model based on defect-related parameters.

[0049] In one specific implementation, such as Figure 3 As shown, a parametric electromagnetic simulation model is constructed based on defect-related parameters, which can be read by a finite-difference time-domain electromagnetic forward modeling solver. During model generation, the material IDs, spatial boundaries, and physical truth values ​​of each object are recorded synchronously. The parametric simulation scene is described using geometric commands, scripted model generation commands, or hierarchical voxel geometry files. Geometric commands use geometric instructions recognizable by the forward modeling solver to directly define objects such as cuboids, cylinders, and thin layers, along with their positions, dimensions, and material properties. Scripted model generation commands use programs such as Python to automatically calculate object boundaries based on sampled parameters and generate geometric description instructions or voxel material files. It is not itself a model file, but rather an implementation method for automatically building the model. The original code generates the distribution of reinforcement, voids, cracks, and aggregates through programs such as loops, polygon filling, and random fields. The hierarchical voxel geometry file divides the simulation region into a regular mesh, with each voxel storing a material ID, which is then mapped to relative permittivity, conductivity, and permeability through a material table.

[0050] Specifically, such as Figure 4 As shown, to improve the exposure and feasibility of complex defect geometry in the FDTD mesh, this embodiment employs voxelized material mapping and sub-voxel coverage processing. A three-dimensional discrete mesh is then established. A material table T is created, ensuring that each material number corresponds to its relative permittivity, conductivity, magnetic permeability, and object category. For reinforcing bars, rust layers, voids, polygonal defects, and polyline cracks, the object region Ωq is first generated in continuous physical space. Then, each Yee mesh element is divided into m×m×m sub-elements, and the coverage of the object within the current element is calculated. (5).

[0051] When multiple objects cover the same mesh cell, the primary material number is written according to the priority of metal or highly conductive targets, rust layer, voids / cracks, aggregate clutter, and concrete matrix; for boundary cells, when the coverage is between 0 and 1 and has not reached the direct write threshold, it is treated as a mixed boundary cell, and the equivalent medium parameters are calculated based on the coverage. (6).

[0052] For polygonal voids, a continuous mask is generated by point-in-polygon determination or boundary scan filling. For polyline cracks, linear cavities or thin voxels of a preset thickness are generated centered on the polyline control points, and isolated elements are deleted through connected component checks. Minimum thickness constraints, connected component preservation, and optional morphological closing operations are applied to the geometric boundaries to maintain the physical continuity of voids, cracks, and corrosion layers in the discrete mesh. The above processing cannot completely eliminate step errors, but it can reduce non-physical scattering caused by single-element spikes, jagged boundaries, and material number jumps.

[0053] In one unrestricted implementation, the voxel model file includes fields such as a 3D data array, a material table, mesh step size, detection surface location, object number mapping, and a random seed. Each element of the data array stores a material number, the material table provides the electromagnetic parameters corresponding to that number, and the object number mapping is used to associate the voxel mesh with the physical ground truth, category, and boundary extent of the defect. The generated voxel file is saved synchronously with the metadata, so subsequent A-scans, B-scans, detection boxes, and segmentation masks can all be traced back to the same physical object.

[0054] The forward simulation in this embodiment is based on Maxwell's curl equation in a lossy medium. For a linear, isotropic, nonmagnetic medium, it can be written as: (7), (8).

[0055] in, For Hamiltonian operators, For curl calculation, For electric field strength, ρ is the magnetic field strength, and ρ is the absolute permittivity of the medium. , The vacuum permittivity is 8.854 × 10⁻⁶. 12 F / m, The relative permittivity of the medium, Permeability, For electrical conductivity, The excitation source current density is used. After discretization using a Yee mesh with central difference, taking the two-dimensional TMz form as an example, An update can be written as: (9), (10). The magnetic field components are updated as follows: (11), (12).

[0056] To ensure numerical stability, the time step must satisfy the Courant condition: (13).

[0057] Where n is the discrete time step number, and i and j are the spatial grid indices in the x and y directions, respectively. , , which represents the staggered positions of the magnetic field components in time or space in the Yee grid. For the nth time step, the grid position The z-direction component of the electric field at that location. For the nth time step, the grid position The z-direction component of the electric field at that location. , These represent the x- and y-direction components of the magnetic field, respectively. , These represent the grid step sizes in two spatial directions; For time step, Excitation source current density The component in the z-direction, and To account for the electric field update coefficient of dielectric conductivity loss, , and These represent the absolute permittivity, conductivity, and permeability of the medium at the current grid location, respectively. Courant stability coefficient; This represents the maximum electromagnetic wave propagation speed in the simulation area.

[0058] Step 2.2: Perform electromagnetic forward modeling and solve the parameterized electromagnetic simulation model to generate A-scan and B-scan data.

[0059] In one specific implementation, such as Figure 6 As shown, a finite-difference time-domain electromagnetic forward modeler is used to solve the parametric electromagnetic simulation model. The finite-difference time-domain electromagnetic forward modeler solves the Maxwell equations for lossy media on the Yee grid, satisfying the spatial step size, time step size, and Courant stability conditions. The simulation domain, materials, source, receiver, boundary conditions, and number of measurement points are set via input files or the model interface.

[0060] It should be noted that an A-scan refers to the sequence of electromagnetic field components recorded by the receiving antenna over time at a fixed measurement point. The horizontal axis represents the sampling time or sampling point number, and the vertical axis represents the received field amplitude. In this embodiment, it is preferable to record the Ez electric field component at the receiving antenna. One A-scan can be represented as: .

[0061] in, This represents the sequence representation of an A-scan, where k is the measurement point number. Let k be the location of the measuring point. Let Δt be the time interval for the nth sampling moment. This represents the number of time sampling points.

[0062] B-scan refers to a two-dimensional matrix formed by arranging the A-scans obtained from each measurement point in the order of the measurement points when the transmitting and receiving antennas move along the same measurement line. Its horizontal axis represents the measurement point number or scanning distance, and its vertical axis represents the sampling time, sampling point number, or equivalent depth after conversion. The matrix elements are the received field amplitude values ​​at the corresponding measurement points and sampling times, and can be represented as: .

[0063] in, The matrix element representation of a B-scan is given, and the measurement point positions satisfy... , This is the starting position of the survey line. This is the step size for measuring points.

[0064] Specifically, the following steps are included: Step 2.2.1: Move the transmitting antenna and receiving antenna sequentially along the preset measurement line, call the time-domain finite-difference electromagnetic forward modeler, and output the received electric field time series as A-scan for each measurement point.

[0065] In one specific implementation, the antenna parameters include waveform type, center frequency, initial position of transmitting antenna, initial position of receiving antenna, transmitting-receiving distance, measurement point step size, time window, and received electric field component; the waveform type is preferably Ricker pulse, and the center frequency is selected within a preset frequency band according to the component thickness, defect size, and detection depth.

[0066] Step 2.2.2: Stack multiple A-scans on the same measurement line into a B-scan matrix according to the measurement point order, and perform device domain degradation, amplitude calibration, background removal, time-varying gain, normalization and image resampling on the B-scan matrix.

[0067] In one specific implementation, device domain degradation includes superimposing Gaussian noise that meets the target signal-to-noise ratio, performing bandpass filtering according to the measured device frequency band, setting a blank row before the zero point of time, cropping or interpolating to the target number of sampling points and measurement points, and performing amplitude calibration according to the amplitude standard deviation of the measured A-scan matrix; B-scan image enhancement in S5 includes median background removal, exponential time-varying gain, robust normalization, hyperbolic tangent grayscale mapping, and preset size resampling.

[0068] This embodiment proposes a device domain alignment process, which converts the forward waveform into a matrix and image with a fixed number of samples, a fixed number of measurement points, a fixed grayscale size, and approximate measured amplitude statistics. This helps to reduce the domain differences between simulation data and measured data.

[0069] Specifically, this embodiment can use an electromagnetic forward modeling solver with FDTD solving capabilities. During simulation, the simulation domain, material, object, Ricker pulse, antenna, receiver, and number of measurement points are set through input files, model interfaces, or voxel model files; the time series of the receiving points is output, and multiple measurement point outputs are merged to form a B-scan matrix of the same measurement line.

[0070] In this embodiment, the measurement point refers to the positional state when the transmitting and receiving antennas are in a certain set of defined locations, completing one electromagnetic transmission and reception sampling. A received electric field time sequence obtained under this state is an A-scan.

[0071] Let the center positions of the transmitting or receiving antennas at the k-th measuring point be respectively and Measurement point center It can be approximated as: (14).

[0072] in, This represents the starting equivalent measurement point position of the survey line, i.e., the lateral coordinate of the first measurement point. This represents the scanning step size between adjacent measurement points.

[0073] Forward solver output number Received electric field sequence at each measuring point This sequence is an A-scan. Stacking N A-scans in the order of measurement points yields a B-scan matrix: (15).

[0074] in, Let be the element in the nth row and kth column of the B-scan matrix, representing the electric field amplitude received by the kth measuring point at time tn. For the number of measurement points, This represents the sampling time corresponding to the nth time sampling point.

[0075] In one embodiment, the target output matrix size is 1697×200, that is, the number of time sampling points is 1697 and the number of measurement points is 200; the B-scan image is resampled into a 640×640 grayscale image.

[0076] Because the ideal FDTD forward modeling results differ from the actual radar equipment acquisition results in amplitude scale, frequency band, noise, and time null point, this embodiment introduces equipment domain constraints on the forward modeling matrix. First, Gaussian noise satisfying the target signal-to-noise ratio is superimposed: (16).

[0077] in, The B-scan matrix after adding noise, This represents Gaussian random noise with a mean of 0, and its second parameter represents the noise variance. in Let be the matrix mean square energy, and SNR be the target signal-to-noise ratio. Then, bandpass filtering is performed on each A-scan to make the frequency band consistent with the measured equipment. (17).

[0078] in, This is the B-scan matrix after bandpass filtering. Represents the first in the matrix A-scan corresponding to each measurement point; Number the measurement points; This indicates a bandpass filtering operation; and These are the low cutoff frequency and high cutoff frequency of the bandpass filter, respectively.

[0079] Then set a blank row for the zero point of time, and calibrate the amplitude based on the standard deviation of the measured original matrix: (18).

[0080] in, To complete the equipment domain matrix after frequency band simulation and amplitude calibration; The standard deviation of the target amplitude of the original matrix of the measured equipment; This represents the standard deviation of the matrix after bandpass filtering.

[0081] Background removal and time-varying gain are applied to the device domain matrix: (19) in, This is the B-scan matrix after background removal; Indicates the sampling time or the location of the time sampling point; Indicates the scanning position on the measurement line; This indicates that the median is calculated along the scanning direction at a fixed sampling time.

[0082] (20).

[0083] in, This is the matrix after time-varying gain processing; It is an exponential time-varying gain coefficient; The maximum gain that can be applied; This indicates that the gain is limited to no more than Within the range.

[0084] Finally, robust normalization and hyperbolic tangent mapping are used to generate grayscale images: (twenty one).

[0085] in, The output B-scan grayscale image; m is the matrix. the median; For matrix Standard deviation; This is the center value of the grayscale. This represents the grayscale mapping amplitude. The hyperbolic tangent mapping scale parameter; (·) is the hyperbolic tangent function; (·,0,255) indicates that the output grayscale value is limited to the range of 0 to 255.

[0086] in, and These represent the matrix midpoint and standard deviation, respectively. , , This refers to the grayscale mean, amplitude, and compression scale. Through this link, the generated B-scan image more closely approximates the measured equipment data in terms of brightness, noise, frequency band, and time zero-point characteristics.

[0087] Each defective object has its center coordinates recorded in physical space. , ),width ,high The class_id and the outer boundary. Let the coordinates of the component detection surface be... The defect depth is: (twenty two).

[0088] Where d is the burial depth of the defect center; Detect the surface coordinates of the component; The physical coordinates of the defect center in the depth direction; This is used to ensure that the calculated burial depth is not less than 0.

[0089] Based on the relative permittivity of concrete Calculate the speed of electromagnetic wave propagation: (twenty three).

[0090] in, The speed at which electromagnetic waves propagate in concrete; The speed of light in a vacuum; is the relative permittivity of concrete.

[0091] It should be noted that this embodiment does not target specific third-party software for protection, but rather protects the complete technical process comprised of engineering structural constraints, physical forward modeling constraints, device domain alignment constraints, voxel material boundary treatment constraints, and physical label mapping constraints. The electromagnetic forward modeling solver can be software, program modules, or hardware-accelerated computing modules capable of finite-difference time-domain or equivalent electromagnetic forward modeling.

[0092] Step 3: Based on the physical truth value, dielectric electromagnetic parameters, antenna position, time zero point, and electromagnetic wave propagation delay relationship, generate physical consistency labeling information corresponding to the A-scan arrival time window, B-scan echo response area, and optional segmentation area, and perform physical consistency quality constraint screening on the physical consistency labeling information to form a labeling sample set.

[0093] Step 3.1: Based on the physical truth value, dielectric electromagnetic parameters, antenna position, time zero point, and electromagnetic wave propagation delay relationship, generate physical consistency labeling information corresponding to the A-scan arrival time window, B-scan echo response area, and optional segmentation area.

[0094] In one specific implementation, the propagation delay relationship includes: calculating the propagation velocity within the concrete based on the relative permittivity of the concrete; calculating the air two-way time based on the air layer distance from the antenna to the component detection surface; calculating the two-way propagation time within the concrete based on the defect depth; and adding the zero-point delay, the air two-way time, and the concrete two-way propagation time to obtain the sampling position of the target on the B-scan time axis; for point-like or line-like targets exhibiting hyperbolic characteristics, constructing the target echo trajectory or its bounding box based on the target's lateral position, target depth, zero-point delay, air layer propagation time, and concrete propagation velocity; the generated annotation information includes at least machine-readable target detection labels, physical truth metadata, A-scan arrival time intervals, sample difficulty levels, and optional segmentation masks, where each annotation is calculated from physical coordinates and device sampling parameters.

[0095] Specifically, such as Figure 5 As shown, let the distance between the antenna and the air layer at the detection surface of the component be... The time delay is 00:00 Then the two-way travel time of the air layer and the arrival time of the target echo are respectively: (twenty four), in, This is the two-way propagation time of the electromagnetic wave in the air layer between the antenna and the component detection surface. The vertical distance from the antenna's equivalent center to the component's detection surface; It is the speed of light in a vacuum.

[0096] (25).

[0097] in, The estimated arrival time of the echo from the defect center; d is the system time zero-point delay; d is the burial depth of the defect center relative to the component inspection surface; This represents the propagation speed of electromagnetic waves in concrete.

[0098] Convert the physical x-axis and arrival time into B-scan normalized labels: (26) in, The normalized coordinates of the defect center in the transverse direction of the B-scan; The physical coordinates of the defect center in the scanning direction; The coordinates of the equivalent measuring point at the start of the survey line; Effective length of the measuring line (27).

[0099] in, The normalized coordinates of the defect center along the B-scan time axis; The time interval between adjacent time sampling points; The total number of sampling points at the target time; This indicates the location of the time sampling point corresponding to the target echo.

[0100] in , For time step, Let be the number of target sampling points. Considering the expansion of the target echo with depth in a B-scan, the normalized width and height can be written as: (28) in, is the normalized width of the target detection box in the horizontal direction of the B-scan; w is the physical width of the defect in the scanning direction; Set a fixed horizontal expansion amount for the target bounding box; The lateral expansion coefficient increases with the depth of the defect. This is the depth-dependent horizontal expansion.

[0101] (29).

[0102] in, The normalized height of the target detection box along the B-scan time axis; The physical height of the defect in the depth direction; This is the minimum equivalent height required to ensure that small-scale defects can be annotated; The amount of sampling points added to the target bounding box along the time axis; This indicates that the larger value between the actual height of the defect and the minimum equivalent height is taken.

[0103] For highly reflective targets such as reinforcing bars, which are approximately point-like or linear, their B-scan trajectories exhibit hyperbolic patterns. This embodiment can further calculate the label trajectory or bounding box using the following formula: (30).

[0104] in, For scanning location The estimated arrival time of the target echo; Let be the lateral coordinate of any equivalent measuring point on the measuring line; The horizontal coordinate of the target center; This is the approximate one-way propagation distance between the measuring point and the target center.

[0105] The above formula determines the label position based on physical space and electromagnetic propagation laws, rather than by subjective selection on the image by humans. For segmentation tasks, simulated geometric voxels can be projected onto a B-scan or equivalent depth map to obtain a mask; for A-scan tasks, the output can include coverage measurement points, arrival sampling point indices, and echo time windows.

[0106] Each sample is named with a unique case_id. The final data files and their organization for a simulation sample in one embodiment include simulation and processing results, physical consistency annotation information, and dataset management information: The original forward modeling matrix, device domain matrix, enhancement matrix, and B-scan image belong to the simulation and processing results; the defect physical truth value, arrival time, target box, or mask in the annotation file and metadata belong to the physical consistency annotation information; the data partitioning file belongs to the dataset management information. The radar image file is used to save the enhanced B-scan grayscale image; the original forward modeling matrix file is used to save the electromagnetic forward modeling matrix; the device domain matrix file is used to save the matrix after noise reduction, filtering, time zero-point calibration, and amplitude calibration; the enhancement matrix file is used to save the matrix after background removal and gain adjustment; the annotation file is used to save machine-readable target detection labels or segmentation masks; the metadata file is used to save sample generation parameters; and the data partitioning file is used to save the training set, validation set, and test set partitioning results.

[0107] Metadata should include at least the sample number, random seed, scene type, and concrete. and The parameters include: thickness of the concrete cover, spacing of the reinforcing bars, radius of the reinforcing bars, physical true value of defects, initial position of the antenna, transmit-receive spacing, measurement step size, time window, zero-point delay, signal processing parameters, tag mapping parameters, and quality inspection results.

[0108] Step 3.2: Perform physical consistency quality constraint screening on the physical consistency annotation information to form an annotation sample set.

[0109] In one specific implementation, the physical consistency quality constraint screening includes checking whether the target is within the coverage area of ​​the survey line, whether the target echo time is within the sampling window, whether the B-scan energy is abnormal, whether the target echo peak is visible, whether the signal-to-noise ratio is lower than the threshold, whether the echo response circumference exceeds the imaging field of view, and whether the sample proportions of rebar, voids, cracks, and corrosion layers meet the preset dataset distribution. The sample set is organized according to sample number into radar images, original forward modeling matrices, equipment domain matrices, enhancement matrices, physical consistency annotation files, metadata, and data partitioning files. The metadata records random seeds, scene types, material parameters, structural parameters, defect physical truth values, scanning parameters, forward modeling solver parameters, signal processing parameters, and label conversion parameters. Samples that fail the screening can be regenerated or marked as difficult samples.

[0110] This embodiment outputs the original forward modeling matrix, device domain matrix, augmentation matrix, radar image, label file, and metadata, and is suitable for target detection, segmentation, temporal classification, two-stream fusion, pre-training, and simulation-to-test transfer learning.

[0111] This embodiment uses random seeds and metadata to record material, structure, scanning, signal processing, and label conversion parameters, enabling sample-level reproduction and algorithm result traceability.

[0112] To better illustrate the superiority of this embodiment, the following simulation example is performed: The simulation domain size is set to approximately 1.0 m × 0.4 m × 0.002 m, with a mesh step size of 0.002 m, and the outer surface coordinates are... The depth was 0.350 m. The relative permittivity of concrete was randomly sampled between 6.0 and 8.5, and the conductivity was randomly sampled between 0.005 S / m and 0.012 S / m; the center frequency was 1.5 GHz; the survey line contained 200 measuring points. Scene types included rebar, voids, cracks, corrosion, honeycombing, cavitation, and mixed defects.

[0113] Figure 7 Simulation samples of mixed scenarios were showcased. Figure 8 For simulation geometric model, Figure 9This is a B-scan radar image. Figure 10 This is a schematic diagram of automatic annotation for a simulated B-scan radar image. The model includes regular rebar, voids, cracks, and aggregate clutter. In the simulated radar image, rebar forms multiple typical hyperbolic echoes, while voids and cracks form localized strong reflections or irregular responses at deeper locations. The tag file records the target bounding boxes in a normalized form of "category x-axis y-axis width and height," for example, category 0 represents rebar, category 1 represents voids, and category 2 represents cracks.

[0114] This embodiment demonstrates that the present invention can simultaneously transform the same physical scene into a geometric model, FDTD waveform, B-scan image, detection label, and metadata, enabling training samples to have a clear physical origin and traceable parameters.

[0115] The system described above can run on a workstation, GPU server, or high-performance computing platform. The method can also be implemented as a computer program, stored in a computer-readable storage medium, and executed by a processor.

[0116] This invention can generate ground-penetrating radar simulation annotation data of internal defects in concrete structures in batches. It can be used for training internal defect identification models of concrete structures, model pre-training, simulation to real-world domain adaptation, algorithm comparison, equipment software verification, and detection strategy optimization. It has clear engineering application value and industrialization prospects.

[0117] Example 2: Embodiment 2 of the present invention provides a concrete internal defect sample generation system based on electromagnetic forward modeling simulation, comprising: The parameter sampling module is used to construct the electromagnetic simulation area of ​​the concrete structure to be tested, and obtain the electromagnetic simulation configuration parameters through constraint sampling, preset assignment and correlation calculation. The forward simulation module is used to construct a parameterized electromagnetic simulation model based on defect-related parameters, and to perform electromagnetic forward simulation and solve the parameterized electromagnetic simulation model to generate A-scan and B-scan data. The physical label generation and filtering module is used to generate physical consistency labeling information corresponding to the A-scan arrival time window, B-scan echo response area, and optional segmentation area based on the physical truth value, medium electromagnetic parameters, antenna position, time zero point, and electromagnetic wave propagation time delay relationship. It also performs physical consistency quality constraint filtering on the physical consistency labeling information to form a labeling sample set.

[0118] Example 3: Embodiment 3 of the present invention provides a computer-readable storage medium storing a computer program adapted for loading by a processor and executing the steps in the method for generating concrete internal defect samples based on electromagnetic forward modeling as described in Embodiment 1 of the present invention.

[0119] Example 4: Embodiment 4 of the present invention provides a computer device, the device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps in the method for generating concrete internal defect samples based on electromagnetic forward modeling as described in Embodiment 1 of the present invention.

[0120] Example 5: Embodiment 5 of the present invention provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the method for generating concrete internal defect samples based on electromagnetic forward modeling as described in Embodiment 1 of the present invention.

[0121] The steps and methods involved in Examples 2, 3, 4 and 5 above correspond to those in Example 1. For specific implementation methods, please refer to the relevant description section of Example 1.

[0122] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0123] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, an optical medium, or a semiconductor medium, etc.

[0124] The above description is only a specific implementation of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application.

Claims

1. A method for generating concrete internal defect samples based on electromagnetic forward modeling, characterized in that, Includes the following steps: An electromagnetic simulation region for the concrete structure to be tested is constructed, and electromagnetic simulation configuration parameters are obtained through constraint sampling, preset assignment, and correlation calculation. Based on the defect-related parameters, a parameterized electromagnetic simulation model is constructed, and electromagnetic forward modeling and solution are performed on the parameterized electromagnetic simulation model to generate A-scan and B-scan data. Based on the physical truth value, medium electromagnetic parameters, antenna position, zero point of time, and electromagnetic wave propagation delay relationship, physical consistency labeling information corresponding to the A-scan arrival time window, B-scan echo response region, and optional segmentation region is generated. The physical consistency labeling information is then filtered by physical consistency quality constraints to form a labeling sample set.

2. The method for generating concrete internal defect samples based on electromagnetic forward modeling as described in claim 1, characterized in that, The specific steps for constructing the electromagnetic simulation region of the concrete structure to be tested are as follows: Establish a coordinate system for the local detection area of ​​the concrete structure to be inspected, and uniformly represent the detection window of the plane, curved or irregular surface as an electromagnetic simulation area in the scanning distance direction, depth direction and thickness or lateral direction.

3. The method for generating concrete internal defect samples based on electromagnetic forward modeling as described in claim 1, characterized in that, Structural constraints include at least the range of component thickness, the range of protective layer thickness, the set of spacing of reinforcing mesh or prestressed tendons, the set of diameters of reinforcing bars or prestressed tendons, the overall initial offset of reinforcing bars, the offset of single bars tied or constructed, the range of survey lines, and optional parameters of construction joints, delamination interfaces, or a priori areas of defects; electromagnetic constraints of materials include at least the sampling range of the relative permittivity and conductivity of concrete, the parameters of ideal conductors or highly conductive materials of reinforcing bars, the dielectric parameters of air, cracks or cavities, and the equivalent dielectric parameters and conductivity of the corrosion layer.

4. The method for generating concrete internal defect samples based on electromagnetic forward modeling as described in claim 1, characterized in that, The parameterized electromagnetic simulation model is solved using a finite-difference time-domain electromagnetic forward modeler. The finite-difference time-domain electromagnetic forward modeler solves the Maxwell equations for lossy media on the Yee grid, satisfying the spatial step size, time step size, and Courant stability conditions.

5. The method for generating concrete internal defect samples based on electromagnetic forward modeling as described in claim 1, characterized in that, The specific steps for performing electromagnetic forward modeling and solving on the parameterized electromagnetic simulation model to generate A-scan and B-scan data are as follows: The transmitting and receiving antennas are moved sequentially along the preset measurement line. The finite-difference time-domain electromagnetic forward modeling solver is invoked to output the time series of the received electric field for each measurement point as an A-scan. Multiple A-scans on the same measurement line are stacked into a B-scan matrix according to the measurement point order. The B-scan matrix is ​​then subjected to device domain degradation, amplitude calibration, background removal, time-varying gain, normalization, and image resampling.

6. The method for generating concrete internal defect samples based on electromagnetic forward modeling as described in claim 1, characterized in that, Physical consistency quality constraint screening includes checking whether the target is within the coverage area of ​​the survey line, whether the target echo time is within the sampling window, whether the B-scan energy is abnormal, whether the target echo peak is visible, whether the signal-to-noise ratio is lower than the threshold, whether the echo response outer region exceeds the imaging field of view, and whether the sample ratio of rebar, voids, cracks and corrosion layers meets the preset dataset distribution.

7. A system for generating concrete internal defect samples based on electromagnetic forward modeling, characterized in that, include: The parameter sampling module is used to construct the electromagnetic simulation area of ​​the concrete structure to be tested, and obtain the electromagnetic simulation configuration parameters through constraint sampling, preset assignment and correlation calculation. The forward simulation module is used to construct a parameterized electromagnetic simulation model based on defect-related parameters, and to perform electromagnetic forward simulation and solve the parameterized electromagnetic simulation model to generate A-scan and B-scan data. The physical label generation and filtering module is used to generate physical consistency labeling information corresponding to the A-scan arrival time window, B-scan echo response area, and optional segmentation area based on the physical truth value, medium electromagnetic parameters, antenna position, time zero point, and electromagnetic wave propagation time delay relationship. It also performs physical consistency quality constraint filtering on the physical consistency labeling information to form a labeling sample set.

8. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method for generating concrete internal defect samples based on electromagnetic forward modeling as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1-6: a method for generating concrete internal defect samples based on electromagnetic forward modeling.

10. A computer device, characterized in that, include: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the method for generating concrete internal defect samples based on electromagnetic forward modeling as described in any one of claims 1-6.

Images