Steel reinforcement cover detection analysis and prediction method and system based on multi-modal data fusion

Abstract

The application provides a reinforced protective layer detection analysis and prediction method and system based on multi-modal data fusion, and relates to the technical field of data processing.The method comprises the following steps: acquiring a multi-modal original data stream, extracting three non-collinear transition points from the multi-modal original data stream, and mapping to a three-dimensional space to construct a space base surface; calculating the center of gravity of the space base surface as a geometric constraint origin and the circumcenter as a space guide pole; calculating the center of gravity and the circumcenter of the space base surface, taking the center of gravity as the geometric constraint origin and the circumcenter as the space guide pole; calculating the spatial dispersion index between the three non-collinear transition points, and combining the spatial distribution characteristics of the space base surface to construct a triangular pyramid shell or a hyperbolic guide cylindrical surface.The application improves the accuracy of reinforced protective layer detection, reduces detection errors, ensures that the protective layer thickness prediction is more in line with the actual working condition, and simultaneously realizes the standardization of the detection process and the visualization of the results.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for detecting, analyzing and predicting the protective layer of reinforcing steel bars based on multimodal data fusion. Background Technology

[0002] In the detection of the protective layer thickness of reinforced concrete structures, commonly used methods include rebar locators based on the principle of electromagnetic induction and reflection wave methods based on ground-penetrating radar. These methods typically process data acquired by a single sensor and estimate the thickness value based on experience or simplified geometric models. In practical engineering, when the rebar arrangement is relatively regular and the concrete medium is relatively homogeneous, these methods can achieve acceptable detection results. However, in certain engineering scenarios, the detection process may encounter some noteworthy situations. For example, when there are multiple layers of rebar, spatial bending sections, or densely intersecting areas of rebar in the structure, single-modal data may sometimes be insufficient to fully reflect the spatial distribution characteristics of the rebar. Furthermore, most detection methods implicitly assume that the protective layer is an ideal homogeneous medium and that the rebar is arranged in a regular straight line. However, in actual components, there may be slight spatial displacement of the rebar, local segregation of the concrete, or changes in surface curvature. These factors may affect the detection signal, causing fluctuations in thickness readings at different locations within the same component.

[0003] For example, during the acceptance inspection of the spiral stirrup section of a bridge pier, the stirrups were arranged in a spatial spiral trajectory, and there was local accumulation of coarse aggregate in the protective layer. When the inspectors used a conventional electromagnetic induction instrument to scan along the circumferential direction, there were obvious differences in the thickness values ​​obtained from different measuring points. If the average or minimum value of multiple measurements is directly taken as the basis for judgment, it may mask the problem of local thinning caused by spatial path distortion, thus affecting the overall judgment of the qualification of the protective layer. Summary of the Invention

[0004] This invention provides a method and system for detecting, analyzing and predicting the protective layer of steel bars based on multimodal data fusion, which solves the problems of reading fluctuations caused by ignoring spatial path distortion and the masking of locally thin areas.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion, the method comprising: Step 1: Obtain the multimodal raw data stream, extract three non-collinear transition points from the multimodal raw data stream, and map them to three-dimensional space to construct a spatial base plane; Step 2: Calculate the centroid and circumcenter of the spatial base plane, take the centroid as the origin of the geometric constraint and the circumcenter as the spatial guiding pole; calculate the spatial dispersion index between the three non-collinear transition points, and construct a triangular pyramidal shell or hyperbolic guiding cylinder based on the spatial distribution characteristics of the spatial base plane. Step 3: Project the triangular pyramidal shell or hyperbolic guiding cylinder onto the spherical coordinate system for normal cutting to obtain nonlinear thickness slices; apply a spatial warping transformation centered on the spatial guiding pole to each nonlinear thickness slice to obtain warped slices; Step 4: Calculate the maximum characteristic direction of the second-order moment of inertia tensor of each warped slice as the principal inertial axis. Divide the warped slice into sub-bands along the principal inertial axis, extract the local centroid of each sub-band and construct the minimum enclosing ellipsoid. Use the center of the ellipsoid as the equivalent centroid to obtain the equivalent centroid sequence. Calculate the ratio of the total length of the line connecting adjacent equivalent centroids to the straight-line distance between the first and last equivalent centroids. Obtain the path distortion factor through hyperbolic cotangent mapping. Step 5: Calculate the initial protective layer thickness value, correct the initial protective layer thickness value using the path distortion factor, and obtain the final protective layer thickness prediction value and the qualification judgment result.

[0006] Secondly, a rebar cover detection, analysis, and prediction system based on multimodal data fusion includes: The spatial datum construction module is used to acquire the multimodal raw data stream, extract three non-collinear transition points from the multimodal raw data stream, and map them to three-dimensional space to construct the spatial datum. The selection module is used to calculate the centroid and circumcenter of the spatial base surface, taking the centroid as the origin of the geometric constraint and the circumcenter as the spatial guiding pole; it calculates the spatial dispersion index between three non-collinear transition points, and constructs a triangular pyramidal shell or hyperbolic guiding cylinder based on the spatial distribution characteristics of the spatial base surface. The warped slice acquisition module is used to project a triangular pyramidal shell or hyperbolic guiding cylinder onto a spherical coordinate system for normal cutting to obtain a nonlinear thickness slice; and to apply a spatial warping transformation centered on the spatial guiding pole to each nonlinear thickness slice to obtain a warped slice. The path distortion factor calculation module is used to calculate the maximum characteristic direction of the second-order moment of inertia tensor of each warped slice as the principal inertial axis. The warped slice is divided into sub-bands along the principal inertial axis, the local centroid of each sub-band is extracted and the minimum enclosing ellipsoid is constructed. The center of the ellipsoid is used as the equivalent centroid to obtain the equivalent centroid sequence. The ratio of the total length of the line connecting adjacent equivalent centroids to the straight-line distance between the first and last equivalent centroids is calculated and the path distortion factor is obtained through hyperbolic cotangent mapping. The result acquisition module is used to calculate the initial protective layer thickness value, correct the initial protective layer thickness value with the path distortion factor, and obtain the final protective layer thickness prediction value and the qualification judgment result.

[0007] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0008] The above-described solution of the present invention has at least the following beneficial effects: This method, by fusing electromagnetic induction signals with multimodal raw data streams of probe displacement and attitude, extracts spatial feature transition points of reinforcing bars and constructs a spatial datum, overcoming the limitation that single-modal data can only reflect local signal features. It can comprehensively characterize the true spatial distribution of reinforcing bars within concrete. Based on the centroid and circumcenter of the spatial datum, geometric constraint origins and spatial guiding poles are established respectively. Combined with the spatial dispersion index, a triangular pyramidal shell or hyperbolic guiding cylinder is adaptively constructed, allowing for matching of the corresponding geometric guiding model according to the actual spatial arrangement characteristics of the reinforcing bars. By forming nonlinear thickness slices through normal cutting in spherical coordinates and applying a spatial warping transformation centered on the spatial guiding poles, the method can restore signal propagation path distortion caused by factors such as inhomogeneity of the concrete medium and changes in surface curvature, weakening the interference of non-ideal conditions such as local segregation and aggregate accumulation on the detection signal, and making the slice features more closely resemble the actual signal propagation laws. By determining the principal inertial axis and dividing it into sub-zones, and extracting the equivalent centroid sequence through the minimum enclosing ellipsoid, the degree of spatial distortion of the rebar's spatial feature points is quantified. The initial thickness value is then corrected using a path distortion factor, resolving issues such as reading fluctuations and the obscuring of locally thin areas caused by neglecting spatial path distortion, thus improving the stability and accuracy of protective layer thickness prediction. A micro-AI agent embedded with sensors and a lightweight AI algorithm engine are used to complete signal processing and thickness calculation, avoiding the risk of misjudgment caused by manual averaging or simple value taking. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the method for detecting, analyzing, and predicting the protective layer of steel bars based on multimodal data fusion, provided by an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of a rebar cover detection, analysis and prediction system based on multimodal data fusion provided in an embodiment of the present invention.

[0011] Figure 3 This is a schematic diagram of multimodal data timestamp alignment and transition point extraction provided by an embodiment of the present invention.

[0012] Figure 4 This is a statistical chart of the extraction accuracy of the equivalent centroid of warped slices provided in an embodiment of the present invention.

[0013] Figure 5 This is a statistical chart comparing the thickness of the protective layer before and after correction, provided in an embodiment of the present invention. Detailed Implementation

[0014] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0015] like Figure 1 As shown, an embodiment of the present invention proposes a method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion. The method includes the following steps: Step 1: Obtain the multimodal raw data stream, extract three non-collinear transition points from the multimodal raw data stream, and map them to three-dimensional space to construct a spatial base plane; Step 2: Calculate the centroid and circumcenter of the spatial base plane, take the centroid as the origin of the geometric constraint and the circumcenter as the spatial guiding pole; calculate the spatial dispersion index between the three non-collinear transition points, and construct a triangular pyramidal shell or hyperbolic guiding cylinder based on the spatial distribution characteristics of the spatial base plane. Step 3: Project the triangular pyramidal shell or hyperbolic guiding cylinder onto the spherical coordinate system for normal cutting to obtain nonlinear thickness slices; apply a spatial warping transformation centered on the spatial guiding pole to each nonlinear thickness slice to obtain warped slices; Step 4: Calculate the maximum characteristic direction of the second-order moment of inertia tensor of each warped slice as the principal inertial axis. Divide the warped slice into sub-bands along the principal inertial axis, extract the local centroid of each sub-band and construct the minimum enclosing ellipsoid. Use the center of the ellipsoid as the equivalent centroid to obtain the equivalent centroid sequence. Calculate the ratio of the total length of the line connecting adjacent equivalent centroids to the straight-line distance between the first and last equivalent centroids. Obtain the path distortion factor through hyperbolic cotangent mapping. Step 5: Calculate the initial protective layer thickness value, correct the initial protective layer thickness value using the path distortion factor, and obtain the final protective layer thickness prediction value and the qualification judgment result.

[0016] In this embodiment of the invention, by fusing electromagnetic induction signals with multimodal raw data streams of probe displacement and attitude, and combining micro-AI agents to extract spatial feature transition points of reinforcing bars and construct spatial datum surfaces, the limitation of single-modal data only reflecting local signal features is overcome, enabling a complete characterization of the true spatial distribution of reinforcing bars within concrete. Geometric constraint origins and spatial guiding poles are established based on the centroid and circumcenter of the spatial datum surface, respectively. Triangular pyramidal shells or hyperbolic guiding cylinders are adaptively constructed using the spatial dispersion index, allowing for matching of the corresponding geometric guiding model according to the actual spatial arrangement characteristics of the reinforcing bars. By forming nonlinear thickness slices through normal cutting in spherical coordinates and applying a spatial warping transformation centered on the spatial guiding poles, the distortion of signal propagation paths caused by factors such as inhomogeneity of the concrete medium and changes in surface curvature can be restored. This weakens the interference of non-ideal working conditions such as local segregation and aggregate accumulation on the detection signal, making the slice features more closely resemble the actual signal propagation laws. By determining the principal inertial axis and dividing it into sub-zones, and extracting the equivalent centroid sequence through the minimum enclosing ellipsoid, the degree of spatial distortion of the rebar's spatial feature points is quantified. The initial thickness value is then corrected using a path distortion factor, resolving issues such as reading fluctuations and the obscuring of locally thin areas caused by neglecting spatial path distortion, thus improving the stability and accuracy of protective layer thickness prediction. A micro-AI agent embedded with sensors and a lightweight AI algorithm engine are used to complete signal processing and thickness calculation, avoiding the risk of misjudgment caused by manual averaging or simple value taking.

[0017] In a preferred embodiment of the present invention, step 1 includes: Step 100: The original induced electromotive force signal is acquired by the electromagnetic sensor installed inside the probe of the rebar cover detector, and the displacement and attitude data of the probe are acquired by the path measurement component inside the host. The original induced electromotive force signal and the displacement and attitude data are sent to the micro-AI agent embedded in the sensor inside the probe for timestamp alignment and spatial coordinate system mapping to generate a multimodal raw data stream, specifically including: At the start of the inspection, the operator activates the rebar cover detector. After power initialization, sensor zero-point calibration, and self-testing of all functional modules, the equipment enters a stable detection state. The transmitter module inside the detector generates a stable alternating excitation current, which is transmitted to the electromagnetic sensor inside the probe. This causes the electromagnetic sensor to generate an alternating electromagnetic field that penetrates the concrete. When this electromagnetic field acts on the internal rebar, the rebar generates a secondary induced magnetic field due to electromagnetic induction. The electromagnetic sensor receives this secondary induced magnetic field and converts it into a continuous induced electromotive force signal, completing the acquisition of electromagnetic modal data. Simultaneously, spatial modal data acquisition is performed, i.e., the path measurement component built into the detector. This component consists of an accelerometer, a gyroscope, a magnetometer, and a height sensor working in concert. The accelerometer collects the linear acceleration of the probe in the three orthogonal directions (X, Y, and Z) in real time. In discrete time series The instantaneous velocity is obtained by numerical integration, i.e.: ; in, , , These are the current times. Velocity in the X, Y, and Z directions; , , They are The velocities in the X, Y, and Z directions at time t; the three-dimensional displacement data are obtained through quadratic numerical integration: ; in, , , These are the current times. Displacement in the X, Y, and Z directions; , , They are Displacement in the X, Y, and Z directions.

[0018] The gyroscope acquires the probe's roll angle, pitch angle, and yaw angle in real time to obtain the rotation angle and rotation rate. The attitude angle is obtained by integrating the angular velocity. ; in, , , These are the current times. The pitch angle, roll angle, and yaw angle are measured. , , These are the current times. The pitch angle, roll angle, and yaw angle are measured. , , These are pitch, roll, and yaw angular velocities, respectively.

[0019] The magnetometer collects the components of the Earth's magnetic field intensity, and calculates the absolute deflection direction of the probe through coordinate system projection. The azimuth angle is... ,in , This represents the horizontal orthogonal component output by the geomagnetic sensor. The height sensor collects the vertical distance between the probe and the concrete surface being measured, obtaining the vertical height data.

[0020] The aforementioned induced electromotive force raw signal, three-dimensional displacement data, rotation angle, deflection direction, and vertical height data are transmitted to the micro AI agent through the device's internal SPI high-speed synchronous interface. The micro AI agent is an embedded edge computing unit embedded inside the probe sensor, consisting of a lightweight AI chip, a data preprocessing unit, a feature extraction unit, an intelligent decision-making unit, and a model storage unit. It is equipped with a lightweight AI algorithm engine that has undergone quantization pruning, knowledge distillation, and tensor decomposition combined compression, and has independent edge processing capabilities. The construction and training process of the lightweight AI algorithm engine is as follows: First, a large multimodal feature extraction model based on Transformer encoders is constructed. The input layer of the model adopts a parallel feature access structure, which can simultaneously accommodate heterogeneous data such as the original induced electromotive force signal, displacement time series, attitude angle series, and azimuth angle series. The middle part of the model consists of 6 stacked Transformer encoders, each layer with an 8-head self-attention structure to capture the temporal abrupt change features and spatial position correlation features of electromagnetic signals. The hidden layer interleaves depth-separable convolutional modules and sparse attention modules to reduce the amount of computation while ensuring feature representation capability. The output layer of the model sets three parallel multi-task learning branches to realize the identification of abrupt change points of electromagnetic signals of steel bars, the mapping of probe local coordinates to global coordinates, and the classification of spatial geometric features of steel bars, forming a pre-trained large model that can complete multimodal data learning and intelligent inference end-to-end.

[0021] A multimodal detection dataset containing no fewer than 50,000 samples was pre-constructed, covering concrete of different strength grades, steel bars of different diameters, different arrangement forms (straight, curved, staggered, multi-layered), and different probe detection postures and scanning paths. All sample data underwent uniform standardization and normalization, and data augmentation operations such as signal noise addition, posture rotation, coordinate translation, and temporal scaling were performed. The AdamW optimizer was used during training, with an initial learning rate of 0.001. A weighted combination of the cross-entropy loss function and the mean squared error loss function was used to obtain the total loss function, where the cross-entropy loss function had a weight of 0.6 and the mean squared error loss function had a weight of 0.4. This total loss function was used to iteratively train the pre-trained large model. Gradient clipping was employed during training to prevent gradient explosion, with a gradient clipping threshold set to 5.0. A cosine annealing learning rate decay strategy was used to gradually reduce the learning rate, with the minimum learning rate decayed to 1× The model then stabilizes and stops decreasing until it converges. When the model achieves a signal mutation point identification accuracy of no less than 98%, an average coordinate mapping error of no more than 0.5 mm, and a steel bar spatial feature classification accuracy of no less than 97% on the independent validation set, it is determined that the model has reached the convergence threshold and pre-training is complete.

[0022] All 32-bit floating-point weights, biases, and activation values ​​in the model are uniformly converted to 8-bit fixed-point representations. While fully preserving the model's core numerical distribution and feature representation capabilities, the parameter storage space is reduced to 1 / 4 of the original 32-bit floating-point parameters, effectively reducing the storage pressure and data retrieval overhead on embedded devices. The weight importance of all feature channels in the convolutional and self-attention layers is ranked, and the mean contribution of all channel weights is calculated. Redundant feature channels with a weight contribution below 20% of the mean are removed, controlling the overall pruning ratio to 50%. Only feature learning paths that play a core role in extracting electromagnetic signals from reinforcing bars and analyzing the spatial geometry of reinforcing bars within concrete are retained, completely eliminating invalid and redundant calculations. Low-rank approximate decomposition is performed on the attention weight matrix of the Transformer structure and the high-dimensional weight matrix of the fully connected layers. Let the original high-dimensional weight matrix have dimension 1. × ,in As the feature input dimension, To define the feature output dimension, the matrix is ​​decomposed into dimensions of... × The left-hand low-rank matrix and dimension are × The right-hand side of the low-rank matrix, where the low-rank dimension The dimension of the original matrix is ​​set to 30%. The feature representation capability of the original high-dimensional weight matrix is ​​approximately restored through matrix multiplication of the left and right low-rank matrices, reducing the amount of matrix multiplication operations and model parameters, and significantly improving the inference speed on embedded devices. A pre-trained large model is used as the teacher model, and a lightweight network with fewer network layers, fewer feature channels, and lower overall computational cost is constructed as the student model. The electromagnetic signal temporal features and rebar spatial location association feature maps output from the intermediate layer of the Transformer encoder in the teacher model, as well as the rebar transition point identification results, spatial coordinate mapping results, and rebar spatial morphology classification inference results from the output layer of the teacher model, are used together as knowledge distillation supervision information. A distillation loss function is constructed by weighting and fusing the mean squared error loss of the intermediate layer feature maps of the teacher and student models and the cross-entropy loss of the output layer inference results. Based on this loss function, the student model is trained and optimized so that the deviation between the student model's output results and the teacher model in key tasks such as induced electromotive force signal processing, rebar transition point extraction, spatial coordinate transformation, and rebar geometric feature calculation is no greater than 2%. After processing with the above four-level joint compression strategy, the model's accuracy in core detection tasks such as rebar signal mutation identification, rebar transition point extraction, and rebar spatial geometry calculation does not decrease significantly, and the overall model size can be reduced to less than 10% of the original pre-trained large model. The lightweight model is then deployed to the micro AI agent to form a lightweight AI algorithm engine, enabling the micro AI agent to have edge-side intelligent capabilities such as multimodal data alignment, spatial coordinate transformation, rebar feature extraction, spatial geometry calculation, and real-time intelligent decision-making. It can independently complete the entire process of multimodal data processing without relying on cloud computing resources.

[0023] The micro-AI agent uses the internal crystal oscillator clock of the detector as the reference clock, assigning synchronous clock identifiers to the electromagnetic sensor, accelerometer, gyroscope, magnetometer, and altitude sensor respectively. For each data stream, clock drift compensation and delay correction are performed based on the inherent transmission delay of the sensor and the acquisition cycle deviation. The clock drift coefficient is set to 1. The timing correction formula is: Corrected standard sampling time = Original sampling time + Inherent transmission delay of the sensor + Clock drift coefficient × Acquisition cycle deviation, where the clock drift coefficient is fixed at 1. The sampling times of all data are aligned to the same point in time of the master clock, ensuring that the electromagnetic signal corresponds to the spatial position at the same moment. After timing alignment, the micro-AI agent performs global three-dimensional spatial coordinate system mapping, that is, pre-establishing a right-handed global three-dimensional spatial coordinate system with the detection starting point as the origin, the initial scanning direction of the probe as the X-axis, the normal of the concrete surface as the Z-axis, and the horizontal and vertical directions as the Y-axis. The probe displacement, attitude, direction, and height data acquired by the path measurement component are converted into spatial coordinates in this global coordinate system through coordinate transformation formula, that is, from the roll angle Pitch angle Yaw angle Construct the attitude rotation matrix The displacement and height of the probe in the local coordinate system are transformed to the global coordinate system using a rotation matrix: ; in , , These are the initial position coordinates. This represents the horizontal displacement component along the probe's scanning direction in the probe's local coordinate system. This represents the horizontal lateral displacement component perpendicular to the scanning direction in the probe's local coordinate system. The vertical height component from the probe to the concrete surface under test in the probe's local coordinate system is used to bind the original induced electromotive force signal to the corresponding spatial coordinates according to the timestamp. This process generates structured data with one-to-one correspondence between signals, coordinates, and time, ultimately producing a multimodal raw data stream that integrates electromagnetic induction signals, displacement information, attitude information, and orientation information.

[0024] Step 101: The micro-AI agent extracts three non-collinear transition points from the multimodal raw data stream. These three non-collinear transition points correspond to the spatial projection points of the starting edge, the central axis, and the ending edge of the same reinforcing bar within the concrete along the direction of the detection probe's movement. Combining displacement and attitude data, these three non-collinear transition points are converted into three spatial coordinate points in a three-dimensional spatial coordinate system, denoted as the first transition point, the second transition point, and the third transition point. Using the first, second, and third transition points as three base points, a spatial base surface spanned by these three base points is constructed, specifically including: The micro-AI agent invokes the internal feature extraction unit to preprocess and calculate the features of the induced electromotive force (EMF) signal in the multimodal raw data stream. Specifically, it first removes environmental noise through sliding filtering, then calculates the signal amplitude of the filtered EMF signal along the probe scanning path. The amplitude of the induced EMF signal is the effective value of the induced EMF signal within the current sampling period, which is obtained by squaring all instantaneous values ​​of the induced EMF within a complete sampling period, summing the sums, dividing the sum by the number of sampling points, and then taking the square root of the result to obtain the amplitude of the induced EMF at the current signal point. Next, it calculates the signal gradient of the filtered induced EMF signal along the probe scanning path, i.e., signal gradient = (amplitude of induced EMF at the current moment - amplitude of induced EMF at the previous moment) / sampling time interval. At the same time, it calculates the abrupt change amplitude and abrupt change slope, where the abrupt change amplitude is the difference between the amplitude of the current signal point and the amplitude of the adjacent previous signal point, and the abrupt change slope is the ratio of the abrupt change amplitude to the corresponding spatial displacement of the probe.

[0025] The location where the mutation amplitude is greater than the preset mutation threshold of 0.8V and the mutation slope reaches an extreme value is identified as the signal mutation point. The preset mutation threshold is determined by calibration and testing on various standard concrete test blocks. The background noise signal in the unreinforced area and the effective induction signal in the reinforced area are collected. The root mean square value of three times the amplitude of the background noise signal is calculated and used as the preset mutation threshold. Assuming the value is 0.8V, the probe spatial position coordinates are correlated and matched to automatically extract three non-collinear spatial transition points. The three transition points correspond to the spatial projection point of the starting edge where the reinforcement first generates a strong induction signal along the probe scanning direction, the spatial projection point of the central axis where the signal amplitude is the largest, and the spatial projection point of the ending edge where the signal amplitude decays to 20% of the initial peak value, completely covering the effective projection range of the reinforcement on the detection path.

[0026] To determine if the three transition points are collinear, calculate the cross product magnitude of the vectors formed by the first and second transition points, and the vectors formed by the second and third transition points. If the magnitude is less than a preset threshold (0.01 mm), the three points are considered collinear. In this case, using the midpoint between the first and third transition points as an auxiliary base point, introduce a small offset (0.5 mm) along the normal direction of the rebar direction to construct a non-collinear virtual third point, replacing the original third transition point, ensuring that the three base points are not collinear. Substitute the probe position, attitude, and orientation data corresponding to the finally determined three transition points (or virtual replacement points) into the same global three-dimensional spatial coordinate system transformation formula as in step 100 to calculate the three-dimensional coordinates of the three transition points in the global coordinate system. Name them as the first transition point, second transition point, and third transition point according to the probe scanning order. After determining the coordinates of the three transition points, the projected length, azimuth angle, horizontal offset, and tilt attitude of the reinforcing bars are calculated using these three coordinates. This eliminates the simplistic assumption of ideal straight-line arrangement of the reinforcing bars in traditional detection methods, and accurately represents the actual spatial position of the reinforcing bars. Using the three non-collinear transition points as spatial base points, a spatial base plane is constructed using a three-point plane equation. That is, let the coordinates of the three transition points be... Calculate the two direction vectors in the plane, namely the first spatial direction vector, from the second transition point. With the first jump point The coordinate difference is obtained, used to characterize the spatial pointing relationship between the two transition points and the second spatial direction vector, which is derived from the third transition point. With the first jump point The coordinate difference is obtained; the normal vector of the spatial base plane is calculated, which is obtained by performing a cross product of the first and second spatial direction vectors. This vector represents the perpendicular direction of the spatial base plane, and finally, the equation of the spatial base plane is obtained. ,in , , The three components of the normal vector determine the spatial orientation of the plane. , , The X, Y, and Z coordinate variables of any point within the spatial base plane in the global three-dimensional spatial coordinate system.

[0027] The spatial base plane bears all the key spatial features of the reinforcing bars along the detection path in a planar form. The extension direction of the reinforcing bars is determined by the first and second spatial direction vectors in the base plane. The tilting posture of the reinforcing bars is characterized by the angle between the base plane normal vector and the vertical direction. The projection range of the reinforcing bars is characterized by the coverage area of ​​the base plane on the detection plane. Thus, the discrete spatial features of the reinforcing bars are unified into a continuous geometric plane, which completely and stably represents the overall spatial projection shape of the reinforcing bars inside the concrete.

[0028] This embodiment solves the problem that single-modal detection data cannot fully reflect the spatial position and attitude information of the reinforcing bar by fusing electromagnetic induction signals with multimodal data of probe displacement and attitude, and completing timestamp alignment and spatial coordinate system mapping through a micro-AI agent. This achieves precise binding between signal features and spatial position. By extracting the three-dimensional transition points corresponding to the starting edge, central axis, and ending edge of the reinforcing bar and constructing a spatial base surface, it can accurately depict the spatial extension shape of a single reinforcing bar along the detection direction. It abandons the simplistic assumption of ideal straight-line arrangement of reinforcing bars in traditional detection, and truly restores the actual spatial projection characteristics of the reinforcing bars inside the concrete, improving the accuracy of describing bent and misaligned reinforcing bars.

[0029] In a preferred embodiment of the present invention, step 2 includes: Step 200: Calculate the centroid of the spatial datum as the origin of the geometric constraint and the circumcenter as the spatial guiding pole; calculate the ratio of the maximum Euclidean distance to the minimum Euclidean distance between the first, second, and third transition points, and define this ratio as the spatial dispersion index, specifically including: The micro-AI agent performs geometric parameter calculations on the constructed spatial base surface. The barycenter coordinates are the arithmetic mean of the coordinates of the three transition points; the barycenter is set as the origin of the geometric constraints. The circumcenter is the center of the circumcircle of the triangle formed by the three transition points, solved using a system of equations based on perpendicular bisectors. This assumes that any point in space is equidistant from the three transition points, satisfying the following condition: ; After expansion and simplification, we obtain the system of linear equations: ; The equations of the spatial datum plane obtained in step 101 are combined to form a system of three linear equations. Solving for the circumcenter yields the circumcenter, which is then designated as the spatial guiding pole. The spatial dispersion index is calculated by first calculating the Euclidean distance between each pair of transition points. The maximum and minimum Euclidean distances are selected from the three sets of distance values. The spatial dispersion index = maximum Euclidean distance / minimum Euclidean distance. This index is used to objectively quantify the degree of dispersion of the three transition points in space. A larger index indicates a more dispersed distribution of transition points, while a smaller index indicates a more regular distribution.

[0030] Step 201: Calculate the ratio of the area of ​​the spatial base surface to the square of the perimeter of the line connecting the three non-collinear transition points. Compare this ratio to the reciprocal of the spatial dispersion index. When the ratio is less than the reciprocal of the spatial dispersion index, construct a triangular pyramidal shell with the point offset by a preset distance from the geometric constraint origin along the normal to the spatial base surface as the apex and the plane containing the three non-collinear transition points as the base. Specifically, this includes: After completing the spatial datum geometry calculation and spatial dispersion index calculation, the micro-AI agent continues to perform the following parameter calculation: calculating the spatial datum area, i.e. Calculate the perimeter of the outline connecting the three transition points, i.e., the sum of the three sets of distance values; calculate the spatial morphological feature ratio, i.e., spatial morphological feature ratio = spatial base area / square of the perimeter of the outline connecting the three transition points; compare the spatial morphological feature ratio with the reciprocal of the spatial dispersion index. When the spatial morphological feature ratio < the reciprocal of the spatial dispersion index, it is determined that the spatial distribution of the three transition points is highly dispersed, and the corresponding reinforcement has complex distribution patterns such as spatial deviation and irregular orientation. At this time, take the geometric constraint origin as the reference point and offset it outward by a preset distance of 10mm along the direction of the spatial base surface normal vector to obtain the cone vertex; take the cone vertex as the cone top and the spatial base surface where the three transition points are located as the base surface, and connect the cone vertex and the three base points in sequence to construct a closed triangular pyramidal shell spatial geometric structure. Let the coordinates of the cone vertex be... Then the equation of the triangular pyramidal shell surface is: In the formula These correspond to three transition points, respectively. Let be the coordinates of any point on the conical shell surface. These are the three-dimensional coordinates of the three transition points that constitute the spatial base. The conical shell is formed by three ruled surfaces, which enables the geometric model to adapt to the spatial form of discretely distributed steel bars.

[0031] Step 202: When the ratio of the area of ​​the spatial base surface to the square of its perimeter is greater than or equal to the reciprocal of the spatial dispersion index, a hyperbolic guiding cylinder is generated by sweeping along the axis using a straight line passing through the origin of the geometric constraint and perpendicular to the spatial base surface as the axis, and using the spatial polygonal line formed by connecting the first, second, and third transition points in sequence as the generatrix. Specifically, this includes: When the ratio of spatial morphological features is greater than or equal to the reciprocal of the spatial dispersion exponent, the spatial distribution of the three transition points is considered relatively regular, and the direction of the reinforcing bars is continuous and smooth. A cylindrical geometric model is used for fitting, i.e., the origin passing through the geometric constraint is selected. Furthermore, the straight line perpendicular to the spatial base plane is taken as the central axis of the model. Let the unit normal vector of the spatial base plane be obtained by the cross product and normalization of the plane spanned by the three transition points, with its direction perpendicular to the spatial base plane. Then the direction vector of the axis is Following the actual scanning sequence of the probe, the first, second, and third transition points are sequentially connected by straight lines to form a continuous and smooth spatial polyline, which serves as the model generatrix. The generatrix parametric equation is as follows: ,in It is the position vector of any point on a polyline in space. It is the relative position along the probe scanning direction (from the beginning to the end of the rebar). At that time, corresponding to the first transition point, When = 0.5, it corresponds to the second transition point. When =1, it corresponds to the third transition point. Using the central axis as the sweep axis, the generatrix is ​​rotated and swept around the axis. A hyperbolic guiding cylinder is generated through parameterized surface equations, i.e., the direction around the axis is set. Rotation angle The rotation matrix is Then the coordinates of any point on the cylinder are: ; in ∈[0, 2π] is the rotation angle parameter. For a unit vector Rotation The standard rotation matrix in radians is obtained from the Rodriguez formula, i.e.: ; In the above formula, It is a 3×3 identity matrix; and These are the sine and cosine values ​​of the rotation angle, respectively. For unit vectors The resulting antisymmetric matrix. Traversal. and The range of values ​​for which a value can be taken can form a complete hyperbolic guiding cylinder.

[0032] This embodiment establishes a stable and unified geometric reference system by determining the geometric constraint origin and spatial guiding poles through the centroid and circumcenter, respectively. Simultaneously, it defines a spatial dispersion index using the Euclidean distance ratio, objectively quantifying the spatial distribution differences of transition points. Based on the comparison results between the spatial datum geometric parameters and the dispersion index, it adaptively selects to construct a triangular pyramidal shell or a hyperbolic guiding cylinder. This allows for matching the corresponding geometric guiding model according to the actual spatial arrangement characteristics of the reinforcing bars, avoiding the adaptation deviation caused by using a single fixed geometric model. This improves the geometric expression accuracy of reinforcing bar characteristics under complex spatial morphology and reduces the interference of inhomogeneity of the internal concrete medium and spatial displacement of the reinforcing bars on detection and analysis.

[0033] In a preferred embodiment of the present invention, step 3 includes: Step 300a involves projecting the triangular pyramidal shell or hyperbolic guiding cylinder onto a spherical coordinate system centered at the geometric constraint origin. Three sampling points are selected at logarithmic intervals along the radial distance. Specifically, this includes projecting the constructed triangular pyramidal shell or hyperbolic guiding cylinder onto a spherical coordinate system centered at the geometric constraint origin, using the line connecting the geometric constraint origin to the spatial guiding pole as the polar axis of the spherical coordinate system, and selecting three sampling points along the radial distance according to a logarithmic interval rule. The formula for calculating the logarithmic interval of the radial sampling points is as follows: = In the formula, It is a natural constant. For the first The radial sampling distance corresponding to each sampling point The base radial distance is taken as the reference distance corresponding to the nominal protective layer thickness of the reinforcing steel. The sampling sequence number is 1, 2, 3, etc. The logarithmic coefficient is set to 0.3. This calculation method ensures that the three sampling points are non-linearly distributed along the polar radius direction.

[0034] Step 301a: Using the sampling points as a reference, construct a series of normal cutting planes; for the triangular pyramid shell, decompose the triangular pyramid shell into several triangular facets, and calculate the intersection line between each normal cutting plane and each triangular facet in turn. That is, determine the intersecting edge by judging that the three vertices of the triangular facet are located on the positive side, negative side, or plane of the cutting plane, respectively, and calculate the coordinates of the intersection point. Connect all intersection points to form the cross-sectional profile of the corresponding cutting plane and the triangular pyramid shell, specifically including: Based on the locations of the three radial sampling points, a series of normal cutting planes perpendicular to the radial direction of the spherical coordinate system are constructed. Each normal cutting plane is perpendicular to the polar radius direction, and its center is located at the polar radius position of the corresponding sampling point. For the case using a triangular pyramidal shell, the overall triangular pyramidal shell constructed in step 201 is first decomposed into three regular triangular facets according to the line structure connecting the cone vertex and the three transition points on the bottom surface, and according to the rule that the interior angle of each facet is not less than 30° and not greater than 120°. The three triangular facets are formed by connecting any two sets of adjacent transition points between the cone vertex and the bottom surface, and the boundaries of the facets completely coincide with the edges of the triangular pyramidal shell. The spatial intersection relationship between each normal cutting plane and each triangular facet is calculated sequentially. First, according to the cutting plane equation, with the corresponding radial sampling point as the center of the plane and perpendicular to the radial direction of the spherical coordinate system, the cutting plane equation is: ; In the formula, The coordinates of any point on the current normal cutting plane are three-dimensional spatial coordinates (obtained by logarithmic interval sampling in step 300a). ) is the unit vector of the current radial direction in the spherical coordinate system. By substituting the coordinates of the three vertices of the triangular facet (all vertices of the triangular pyramid shell in step 201, including the cone vertex and the three transition points) into the above cutting plane equation, if the result is greater than 0, it means that the vertex is located on the positive side of the cutting plane; if the result is less than 0, it means that it is located on the negative side; if the result is equal to 0, it means that the vertex is located on the cutting plane. Based on the position of the three vertices, determine the edge of the facet that intersects the cutting plane (i.e., the edge where the two vertices are located on the positive and negative sides of the plane, respectively); then use the spatial linear interpolation formula to calculate the coordinates of the intersection point, i.e., the intersection point coordinates = In the formula, , The coordinates of the two endpoints of the intersecting edges of the facets (the endpoints of the triangular pyramid shell in step 201). These are the interpolation coefficients. = Where Da is the endpoint Substituting the result into the left side of the equation for the cutting plane, where Db is the endpoint... Substitute the calculation result obtained by substituting into the left side of the cutting plane equation. Connect all the calculated intersection points in the same normal cutting plane in sequence to form the closed cross-sectional profile formed by the intersection of the cutting plane and the triangular pyramid shell.

[0035] Step 302a: For the hyperbolic guiding cylinder, the hyperbolic guiding cylinder is parameterized as a sweeping surface of a hyperbola along the axis. The cross-sectional boundary is obtained by solving for the intersection points of the cutting plane and the hyperbolic parametric curve. The spatial region enclosed by the cross-sectional profiles obtained from all normal cutting planes is defined as a nonlinear thickness slice, resulting in three nonlinear thickness slices with logarithmic changes in thickness along the polar radius direction, specifically including: For the case using a hyperbolic guide cylinder, the hyperbolic guide cylinder and its parameterized sweep equation, constructed in step 202, are used to uniformly describe the cylinder as a continuous spatial surface formed by a hyperbolic contour line sweeping and extending at a constant speed around the central axis, ensuring that the spatial shape of the cylinder is consistent with the previously constructed geometric model. The micro-AI agent solves the cutting plane equation (i.e., the normal cutting plane constructed based on spherical coordinate projection, perpendicular to the polar radius direction) constructed in step 301a simultaneously with the parameterized equation of the hyperbolic guide cylinder, calculating the spatial intersection coordinates of each normal cutting plane and the parameter curve of the hyperbolic guide cylinder. After the solution is completed, all continuous intersection points are smoothly connected sequentially around the central axis of the hyperbolic guide cylinder to form a closed and continuous cross-sectional boundary curve, ensuring that the boundary has no breaks or overlaps, accurately reflecting the cross-sectional shape of the hyperbolic guide cylinder at the cutting plane. The closed spatial region enclosed by the cross-sectional profiles obtained by intersecting all normal cutting planes with the hyperbolic guide column surface is uniformly defined as a nonlinear thickness slice. Combined with the sampling requirement of logarithmic interval, three nonlinear thickness slices with thicknesses varying logarithmically along the polar radius direction of the spherical coordinate system are finally obtained. The thickness variation of each slice is adapted to the polar radius extension direction, which fits the actual needs of rebar inspection.

[0036] Step 300b: For each nonlinear thickness slice, apply a spatial warping transformation. The kernel function of the spatial warping transformation is a Gaussian radial basis function centered at the spatial guiding pole. In the spatial warping transformation, each point on the current nonlinear thickness slice generates a nonlinear radial displacement based on its Euclidean distance from the spatial guiding pole. The magnitude of the radial displacement follows a Gaussian distribution, such that points closer to the spatial guiding pole have larger displacements, and points farther away have smaller displacements. Each nonlinear thickness slice after the spatial warping transformation is denoted as a warped slice, resulting in three warped slices, specifically including: A spatial warping transformation is applied to each nonlinear thickness slice. This spatial warping transformation uses a Gaussian radial basis function centered at the spatial guiding pole as the transformation kernel function. The kernel function is calculated using the following formula: =exp(- ), where Gaussian weights, The distance is the Euclidean distance from the current point on the slice to the spatial guiding pole. The preset spatial warping control radius is set to 20% of the maximum span along the radial direction of the current slice. During the spatial warping transformation, for each spatial point on the nonlinear thickness slice, the Euclidean distance between it and the spatial guiding pole must first be calculated. Then, the Gaussian weight corresponding to that point is calculated according to the kernel function formula. Based on this Gaussian weight, the nonlinear radial displacement of that point is calculated, i.e., nonlinear radial displacement = Gaussian weight × preset maximum displacement, where the preset maximum displacement is set to 5% of the nominal protective layer thickness of the rebar (e.g., when the nominal thickness is 20mm, the preset maximum displacement = 1mm). The magnitude of the nonlinear radial displacement is positively correlated with the Gaussian weight; the larger the Gaussian weight, the larger the nonlinear radial displacement. This leads to a pattern where the closer to the spatial guiding pole, the larger the Gaussian weight and the larger the nonlinear radial displacement; the farther away, the smaller the Gaussian weight and the smaller the nonlinear radial displacement. Overall, it exhibits a nonlinear deformation characteristic of strong at the center and weak at the periphery. Each nonlinear thickness slice after spatial warping transformation is denoted as a warped slice. The above warping transformation operation is performed on the three nonlinear thickness slices in sequence to finally obtain a set of warped slices that are logarithmically distributed along the polar radius and have spatial nonlinear deformation characteristics.

[0037] This embodiment employs spherical coordinate projection and logarithmic interval sampling, enabling the slice thickness to adaptively vary along the detection radial direction. This better reflects the signal attenuation and spatial distribution patterns of reinforcing bars within concrete, improving the realism of thickness calculations. For triangular pyramidal shells and hyperbolic guiding cylinders, patch intersection calculations and parametric equation solutions are used respectively, ensuring the stable generation of closed and continuous cross-sectional profiles under both regular and irregular reinforcing bar orientations, avoiding slice breakage or distortion. Based on the Gaussian radial basis warp transform of the spatial guiding poles, the spatial distortion effects caused by internal concrete stress distribution and medium inhomogeneity can be simulated, making the slice shape more closely resemble the actual engineering environment and improving the accuracy of reinforcing bar positioning, protective layer thickness calculation, and spatial shape reconstruction.

[0038] In a preferred embodiment of the present invention, step 4 includes: Step 400a: For each warped slice, calculate the mean coordinates of all points on the corresponding warped slice as the centroid. For each point, calculate the offset vector relative to the centroid. Based on the offset vector, determine the six independent components of the second-order moment of inertia tensor, forming a third-order symmetric tensor. Solve for the eigenvalues ​​of the third-order symmetric tensor, and take the eigendirection corresponding to the largest eigenvalue. Define the corresponding eigendirection as the principal inertial axis, specifically including: For each warped slice that has undergone spatial warping transformation, the mean coordinates of all points on the slice are calculated one by one. This mean coordinate is used as the centroid of the warped slice to ensure that the centroid position accurately reflects the overall spatial position of the slice. For each point on the warped slice, the difference between the point's coordinates and the centroid coordinates is calculated to obtain the offset vector of each point relative to the centroid, i.e., offset vector = current point coordinates - centroid coordinates. This offset vector reflects the spatial position deviation of each point relative to the centroid. Based on the offset vectors of all points, the six independent components of the second-order moment of inertia tensor are determined, forming a third-order symmetric tensor. With the centroid as the local coordinate origin, the six independent components of the second-order moment of inertia tensor are calculated based on the coordinate components of all offset vectors, thus constructing a third-order symmetric tensor. The calculation formulas for each component are as follows: ; ; ; ; ; ; In the formula The first The offset components of a point relative to the centroid on the three coordinate axes of the local coordinate system; , , These are the moments of inertia of the three principal diagonal axes; , , These are three non-diagonal intersecting moments of inertia.

[0039] Arranging the above six components in a symmetrical manner forms a third-order symmetric moment of inertia tensor: ; The symmetric tensor is decomposed into eigenvalues ​​using the characteristic equation method, i.e., the characteristic equation is constructed first. ,in For the eigenvalues ​​to be found, Given a third-order identity matrix, its expansion yields a cubic algebraic equation, which is then solved to obtain three eigenvalues. These eigenvalues ​​are compared, and the largest eigenvalue is selected. This largest eigenvalue is then substituted back into the system of linear homogeneous equations. In the process, the corresponding non-zero eigenvectors are obtained by solving. The spatial direction pointed to by the feature vector is defined as the principal inertial axis of the current warped slice. The principal inertial axis represents the dominant extension direction of the spatial points of the slice.

[0040] Step 401a: Divide the current warped slice into six sub-zones along the principal inertial axis, calculate the local barycenter coordinates of all points within each sub-zone, and obtain six local barycenter points; construct a minimum enclosing ellipsoid using the six local barycenter points, calculate the center coordinates of the minimum enclosing ellipsoid, and use the center coordinates as the equivalent barycenter of the current warped slice; traverse all three warped slices to obtain an equivalent barycenter sequence consisting of three equivalent barycenters, specifically including: After determining the principal inertial axis, the current warped slice is evenly divided into six sub-zones along the principal inertial axis. The width of a single sub-zone is determined based on the total length of the current warped slice along the principal inertial axis (i.e., one-sixth of the total length). The number of spatial points contained in each sub-zone is controlled within the range of 30 to 50 to ensure that the width of each sub-zone is consistent and the distribution of points is balanced, avoiding situations where the points within the sub-zone are too dense or too sparse. For each sub-zone, the mean coordinates of all points within the sub-zone are calculated to obtain the local centroid coordinates of each sub-zone, resulting in six local centroids. Based on these six local centroids, a minimum bounding ellipsoid that can completely enclose all local centroids is constructed. The specific construction process is as follows: First, the three-dimensional coordinates of the six local centroids are collected to determine the specific position of each local centroid in space. The mean coordinates of the six local centroids are calculated to obtain the overall centroid coordinates. ( , , This is used as the initial reference point for the center of the ellipsoid. The distance from each local centroid to the center of the ellipsoid is calculated. The straight-line distances are taken, and the maximum value is taken as the semi-major axis of the ellipsoid. ; Calculate the distance from each local centroid to the ellipsoid center. The straight-line distances are taken as 1.2 times the minimum value, which is used as the minor semi-axis of the ellipsoid. .

[0041] Substitute the coordinates of the six local centroids into the ellipsoid equation. =1 (Because the ellipsoid is symmetrical about the direction of the rebar detection, the minor semi-axis is the same as the semi-axis perpendicular to the detection direction, both being 1) ),in, Let the coordinates be the three-dimensional coordinates of any point on the ellipsoid, if substituting all local centroids satisfies... If the value is ≤1, the parameter is reasonable; if there are points outside the range, then the major semi-axis should be adjusted. Expand by 5% and re-verify until all points are inside the ellipsoid. The equivalent centroid can comprehensively fit and balance the spatial positions of the local centroids of all sub-zones of the warped slice, taking into account the distribution characteristics of the local centroids of each sub-zone. It reflects the overall spatial position, extension direction, and distribution pattern of the warped slice, thus comprehensively reflecting the overall spatial position of the warped slice and avoiding overall positioning errors caused by deviations in the local centroids of individual sub-zones, ensuring positioning accuracy. Following the same calculation method, the equivalent centroids of the other two warped slices are calculated sequentially to obtain three equivalent centroid coordinates. These three equivalent centroids are arranged in order to form an equivalent centroid sequence.

[0042] Step 400b: Connect two adjacent equivalent barycenters in the equivalent barycenter sequence topologically to obtain two spatial lines; calculate the total length of the two spatial lines and the straight-line distance between the first and last equivalent barycenters in the equivalent barycenter sequence; calculate the ratio of the total length of the spatial lines to the straight-line distance to obtain an initial ratio; apply a hyperbolic cotangent mapping to the initial ratio, and use the mapped result as the path distortion factor, specifically including: A topological connection is performed on the equivalent centroid sequence. Adjacent equivalent centroids in the sequence are connected by lines, forming two spatial lines. The lengths of these two spatial lines are calculated separately and then added together to obtain the total length of the spatial lines. Simultaneously, the straight-line distance between the first and last equivalent centroids in the sequence is calculated. This distance represents the shortest path between the two points, visually reflecting the linear distribution of the equivalent centroid sequence. The ratio of the total length of the spatial lines to this straight-line distance is calculated to obtain an initial ratio. This initial ratio directly reflects the curvature of the equivalent centroid sequence. If the equivalent centroid sequence is linearly distributed, the total length of the spatial lines and the straight-line distance are approximately equal, and the initial ratio is close to 1, indicating a small degree of curvature and a relatively regular spatial shape of the reinforcing bars. If the equivalent centroid sequence exhibits significant curvature, the total length of the spatial lines will be greater than the straight-line distance, and the initial ratio will be greater than 1. A larger ratio indicates a more pronounced degree of curvature in the equivalent centroid sequence, meaning a more irregular spatial shape of the reinforcing bars.

[0043] Apply a hyperbolic cotangent mapping to the initial ratio, i.e., the path distortion factor = = , It is a natural constant, due to the initial ratio ≥1, the range of the path distortion factor is The path distortion factor is positively correlated with the curvature of the equivalent centroid sequence. The larger the path distortion factor, the greater the curvature of the equivalent centroid sequence and the more irregular the spatial shape of the reinforcing bars. The closer the path distortion factor is to 1, the closer the equivalent centroid sequence is to a straight line distribution and the more regular the spatial shape of the corresponding reinforcing bars.

[0044] This embodiment captures the spatial distribution characteristics of warped slices by calculating the centroid, offset vector, and second-order moment of inertia tensor. The determination of the principal inertial axis provides a clear benchmark for subsequent slice segmentation and parameter calculation, avoiding directional deviations during slice processing. By dividing into sub-zones and calculating local and equivalent centroids, irregular warped slices can be normalized. The construction of the minimum enclosing ellipsoid accurately reflects the overall spatial position of the slices, and the acquisition of the equivalent centroid sequence provides reliable data support for calculating the path distortion factor. By calculating the path distortion factor, the spatial curvature and distribution of the reinforcing bars can be quantitatively reflected, solving the problem of neglecting the spatial deformation of the reinforcing bars, which leads to deviations in the calculation of the protective layer thickness.

[0045] In a preferred embodiment of the present invention, step 5 includes: Step 500: The lightweight AI algorithm engine embedded in the micro-AI intelligent body within the probe performs a sliding window integration on the original induced electromotive force signal to obtain an energy decay curve. The time corresponding to the peak position in the energy decay curve is extracted as the arrival time of the reflected echo from the rebar. Simultaneously, combined with the real-time position coordinates of the probe in the displacement and attitude data, and using the pre-calibrated electromagnetic induction signal amplitude-distance relationship curve, the initial protective layer thickness value is calculated. Specifically, this includes: The micro-AI agent embedded in the sensor within the probe is activated. This agent integrates a lightweight AI algorithm engine. First, it performs noise filtering on the raw induced electromotive force signal acquired by the probe using an adaptive filtering algorithm (high-frequency noise is controlled below 50Hz, and environmental interference signals are controlled within ±0.5mV to ensure that interference signals do not affect subsequent processing). After filtering, effective features are extracted from the filtered raw signal to remove invalid interference data. A time-alignment algorithm is then used to calibrate the extracted signal, matching the signal acquisition time with the probe position data based on the timestamp acquired by the probe's built-in sensor. This corrects timing deviations during signal transmission, ensuring accurate spatial location and time point correspondence for each signal. Finally, a sliding window integration method is used to calculate the preprocessed induced electromotive force signal. The sliding window length is 100ms, the sliding step size is 10ms, and the sampling frequency is 100Hz. The specific formula for the integration calculation is as follows: ; in, This is the preprocessed induced electromotive force signal. The starting time of integration, The integration time interval (i.e., the sliding window step size of 10ms) is used. For the first Each sampling time, The total number of samples within the integration interval is given by this formula. The integral result obtained by this formula is the core data of the energy decay curve, which clearly shows the energy change pattern during the propagation of electromagnetic waves.

[0046] The peak amplitude is extracted from the energy decay curve. This peak amplitude is negatively correlated with the thickness of the concrete cover (the closer the distance, the stronger the induced electromotive force). Combined with the amplitude-thickness correspondence pre-calibrated on standard concrete test blocks, the initial concrete cover thickness is calculated. The calibration and calculation process is as follows: Before testing, using the same concrete cover testing instrument, select no fewer than 5 groups of standard concrete test blocks with different nominal thicknesses (e.g., 10mm, 15mm, 20mm, 25mm, 30mm). Each thickness test block is tested 10 times, and the peak amplitude of each test is recorded. The arithmetic mean is taken as the calibration amplitude corresponding to that thickness. The calibration thickness value is then... With the calibration amplitude ( =1 to 5), perform least squares fitting to obtain the amplitude-thickness relationship in exponential decay form: = ,in, The thickness of the protective layer to be determined (unit: mm). The measured peak amplitude is expressed in mV. This represents the theoretical maximum amplitude when the probe is in close contact with the surface of the rebar (distance is 0). This is the attenuation coefficient. and The fitting formula is obtained by fitting the calibration data. ,right and For linear regression, the intercept is and the slope is . － .like The maximum calibrated amplitude (85mV) is then taken as follows. =85mV; if If the minimum calibration amplitude is less than 25mV, then take... =25mV; otherwise, take the measured value. After calculating the initial protective layer thickness value, combine the real-time position coordinates of the probe in the displacement and attitude data to perform spatial consistency verification on the thickness value. If the coefficient of variation (ratio of standard deviation to mean) of the thickness value of three consecutive sampling points exceeds 15%, or the change of probe attitude angle (pitch angle, roll angle) exceeds 5°, it is determined that the data segment is affected by probe tilt or off-ground interference, and is discarded and prompted for retesting.

[0047] Step 501: Multiply the initial protective layer thickness value with the path distortion factor to obtain the corrected protective layer thickness prediction value; compare the corrected protective layer thickness prediction value with a preset pass / fail threshold; if it is greater than or equal to the pass / fail threshold, it is judged as passable; otherwise, it is judged as failable, and the corrected protective layer thickness prediction value and pass / fail judgment result are output through the host display screen, specifically including: The path distortion factor is multiplied by the initial protective layer thickness value to obtain the corrected predicted protective layer thickness value, i.e., corrected predicted protective layer thickness value = initial protective layer thickness value × path distortion factor. A larger path distortion factor indicates more significant spatial bending and attitude deviation of the rebar, resulting in a longer electromagnetic wave propagation path. The initial protective layer thickness calculated directly based on straight-line propagation will be underestimated. By multiplying by a path distortion factor greater than 1 (the larger the factor for more severe bending), the underestimated initial thickness is reasonably amplified and compensated, thus offsetting systematic errors caused by factors such as uneven concrete medium, rebar spatial deviation, and non-linear propagation paths, making the calculation result closer to the true thickness of the rebar protective layer. The corrected predicted protective layer thickness value is compared with a preset acceptable threshold, which is set according to the rebar testing industry standard and is 15mm (corresponding to the standard for conventional rebar protective layer acceptance; if the testing scenario is a special component, the threshold can be adjusted according to actual needs). If the corrected predicted protective layer thickness value is ≥15mm, the rebar protective layer thickness is considered acceptable; if the corrected predicted protective layer thickness value is <15mm, the rebar protective layer thickness is considered unacceptable. After the judgment is completed, the corrected predicted value of the protective layer thickness and the corresponding qualification judgment result are simultaneously output through the host display screen.

[0048] This embodiment employs a sliding window integration process to process the induced electromotive force signal, effectively filtering signal noise, extracting the arrival time of reflected echoes, and calculating the initial protective layer thickness by combining it with the electromagnetic wave propagation laws, ensuring the accuracy of the initial data. Correcting the initial protective layer thickness using a path distortion factor can offset detection errors caused by factors such as uneven concrete media and rebar spatial misalignment, making the corrected predicted protective layer thickness closer to the actual situation and improving detection accuracy. Through conformity assessment and result output, staff can quickly determine whether the rebar protective layer thickness meets the standards, clarify the detection results, and avoid potential engineering hazards caused by detection errors.

[0049] like Figure 2 As shown, embodiments of the present invention also provide a rebar cover detection, analysis, and prediction system based on multimodal data fusion, comprising: The spatial datum construction module is used to acquire the multimodal raw data stream, extract three transition points from the multimodal raw data stream, and map them to three-dimensional space to construct the spatial datum. The selection module is used to calculate the centroid and circumcenter of the spatial base surface, taking the centroid as the origin of the geometric constraint and the circumcenter as the spatial guiding pole; it calculates the spatial dispersion index between three non-collinear transition points, and constructs a triangular pyramidal shell or hyperbolic guiding cylinder based on the spatial distribution characteristics of the spatial base surface. The warped slice acquisition module is used to project a triangular pyramidal shell or hyperbolic guiding cylinder onto a spherical coordinate system for normal cutting to obtain a nonlinear thickness slice; and to apply a spatial warping transformation centered on the spatial guiding pole to each nonlinear thickness slice to obtain a warped slice. The path distortion factor calculation module is used to calculate the maximum characteristic direction of the second-order moment of inertia tensor of each warped slice as the principal inertial axis. The warped slice is divided into sub-bands along the principal inertial axis, the local centroid of each sub-band is extracted and the minimum enclosing ellipsoid is constructed. The center of the ellipsoid is used as the equivalent centroid to obtain the equivalent centroid sequence. The ratio of the total length of the line connecting adjacent equivalent centroids to the straight-line distance between the first and last equivalent centroids is calculated and the path distortion factor is obtained through hyperbolic cotangent mapping. The result acquisition module is used to calculate the initial protective layer thickness value, correct the initial protective layer thickness value with the path distortion factor, and obtain the final protective layer thickness prediction value and the qualification judgment result.

[0050] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0051] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0052] Experimental example: I. Experiment Overview This embodiment addresses the problem of detecting the concrete cover thickness of prestressed concrete box girders in a highway bridge project. A verification experiment is conducted using a detection, analysis, and prediction method based on multimodal data fusion. The experimental object is a three-span continuous box girder structure, with a single span of 35m. The box girder has a top slab thickness of 280mm, a web thickness of 400mm, and a bottom slab thickness of 250mm. The concrete design strength grade is C50, and the designed concrete cover thickness is 25mm (top slab), 30mm (web), and 20mm (bottom slab).

[0053] The experiment employed a dual-mode detection system combining electromagnetic induction and radar waves to collect data on the concrete cover of reinforcement from eight typical sections of a prestressed box girder. The detection sections were located at mid-span, L / 4, and supports, with 12 detection points at each section, totaling 96 detection points. Simultaneously, 12 core samples were taken for verification, serving as a reference benchmark for the true thickness of the concrete cover.

[0054] II. Experimental Conditions The experimental conditions are shown in Table 1.

[0055] Table 1 Summary of Experimental Conditions

[0056] III. Experimental Procedures and Results Step 1: Multimodal data acquisition and timestamp alignment: Two detection methods, electromagnetic induction and radar waves, were used to simultaneously acquire data on the concrete cover of the reinforcing steel at eight cross-sections. The sampling rate of the electromagnetic induction signal was 10Hz, and the sampling rate of the radar wave was 5Hz. Due to the different response delays of the two sensors, the electromagnetic induction signal led the radar wave signal by an average of approximately 12.5ms. Timestamp alignment was performed on the two modal data using an alignment strategy based on transition point feature matching. First, signal transition points were detected within a sliding time window, and the transition time and amplitude features were extracted. Then, using the electromagnetic induction signal as a reference, the time offset of the radar wave signal was calculated and corrected. After alignment, the timing deviation between the two modal data decreased from an average of 12.5ms to 1.8ms, improving the alignment accuracy by 85.6%. The average signal-to-noise ratio (SNR) of the extracted transition points was 28.3dB, with 92% of the sampling points having an SNR exceeding the 25dB threshold. The experimental results are shown in Table 2.

[0057] Table 2. Timestamp Alignment Results of Multimodal Data ; like Figure 3 As shown, the timing deviation of each sampling point before alignment fluctuated in the range of 5 to 25 ms, and the deviation converged to the range of 1 to 3 ms after alignment. At the same time, the signal-to-noise ratio of the transition point extraction exceeded the 25 dB threshold in most sampling points, indicating that the transition point feature matching strategy effectively improved the data alignment quality.

[0058] Step 4: Equivalent centroid extraction and protective layer thickness correction: Within each subdomain of the warped slice, the equivalent centroid is extracted using the minimum bounding ellipsoid method, serving as a reference point for correcting the protective layer thickness. The location of the equivalent centroid is compared with that of the direct centroid method and the equidistant segmentation method to verify the superior positioning accuracy of the minimum bounding ellipsoid method. Experimental results show that the minimum bounding ellipsoid method used in this invention has an average positioning error of 0.12 mm across the six sub-bands, a 72.1% reduction compared to the direct centroid method (0.43 mm) and a 61.3% reduction compared to the equidistant segmentation method (0.31 mm). This improved positioning accuracy directly enhances the correction effect on the protective layer thickness. The experimental results are shown in Table 4.

[0059] Table 4 Comparison of Equivalent Centroid Extraction Accuracy ; like Figure 4 As shown, the present invention achieves the lowest positioning error in all six sub-strips, and the error fluctuation range is the smallest (0.09 to 0.15 mm), indicating that the minimum enclosing ellipsoid method has better robustness in complex reinforcement distribution environments.

[0060] The protective layer thickness test results after equivalent centroid correction are shown in Table 5. Among the 12 core sampling verification points, the average absolute error between the measured value and the actual value before correction was 3.58 mm, which was reduced by 0.42 mm after correction, showing a significant correction effect. Among them, point T-05 (bottom plate, designed thickness 20 mm) had the largest error before correction (2.3 mm), which was reduced to 0.3 mm after correction.

[0061] Table 6. Results of protective layer thickness testing before and after correction.

[0062] like Figure 5 As shown, the corrected detection values ​​closely match the actual values ​​(green bars), while the uncorrected detection values ​​(red bars) are systematically too small. After the equivalent centroid correction, the protective layer thickness at all detection points reaches or exceeds the design acceptable threshold (15mm), demonstrating a clear and significant correction effect.

[0063] IV. Experimental Conclusions The alignment method based on transition point feature matching reduces the time deviation between electromagnetic induction and radar wave data from an average of 12.5ms to 1.8ms, improves the alignment accuracy by 85.6%, and achieves a transition point detection rate of 95.6%.

[0064] The average positioning error of the six sub-bands was 0.12 mm, which was 72.1% lower than that of the direct centroid method and 61.3% lower than that of the equidistant segmentation method. After correction, the protective layer thickness MAE decreased from 3.58 mm to 0.42 mm, showing a significant correction effect.

[0065] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion, characterized in that, The method includes: Step 1: Obtain the multimodal raw data stream, extract three non-collinear transition points from the multimodal raw data stream, and map them to three-dimensional space to construct a spatial base plane; Step 2: Calculate the centroid and circumcenter of the spatial base plane, take the centroid as the origin of the geometric constraint and the circumcenter as the spatial guiding pole; calculate the spatial dispersion index between the three non-collinear transition points, and construct a triangular pyramidal shell or hyperbolic guiding cylinder based on the spatial distribution characteristics of the spatial base plane. Step 3: Project the triangular pyramidal shell or hyperbolic guiding cylinder onto the spherical coordinate system for normal cutting to obtain nonlinear thickness slices; apply a spatial warping transformation centered on the spatial guiding pole to each nonlinear thickness slice to obtain warped slices; Step 4: Calculate the maximum characteristic direction of the second-order moment of inertia tensor of each warped slice as the principal inertial axis. Divide the warped slice into sub-bands along the principal inertial axis, extract the local centroid of each sub-band and construct the minimum enclosing ellipsoid. Use the center of the ellipsoid as the equivalent centroid to obtain the equivalent centroid sequence. Calculate the ratio of the total length of the line connecting adjacent equivalent centroids to the straight-line distance between the first and last equivalent centroids. Obtain the path distortion factor through hyperbolic cotangent mapping. Step 5: Calculate the initial protective layer thickness value, correct the initial protective layer thickness value using the path distortion factor, and obtain the final protective layer thickness prediction value and the qualification judgment result.

2. The method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion according to claim 1, characterized in that, Step 1 includes: The original induced electromotive force signal is collected by the electromagnetic sensor installed inside the probe of the rebar cover detector, and the displacement and attitude data of the probe are collected by the path measurement component inside the host. The original induced electromotive force signal and displacement and attitude data are sent to the micro AI agent embedded in the sensor inside the probe for timestamp alignment and spatial coordinate system mapping to generate a multimodal original data stream. The micro-AI agent extracts three non-collinear transition points from the multimodal raw data stream. These three non-collinear transition points correspond to the spatial projection points of the starting edge, the central axis, and the ending edge of the same steel bar inside the concrete along the moving direction of the detection probe. Combining displacement and attitude data, the three non-collinear transition points are converted into three spatial coordinate points in a three-dimensional spatial coordinate system, which are denoted as the first transition point, the second transition point, and the third transition point, respectively. Using the first transition point, the second transition point, and the third transition point as three base points, a spatial base surface spanned by the three base points is constructed.

3. The method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion according to claim 2, characterized in that, Calculate the spatial dispersion index between three non-collinear transition points, and construct a triangular pyramidal shell or hyperbolic guiding cylinder based on the spatial distribution characteristics of the spatial datum, including: Calculate the ratio of the maximum Euclidean distance to the minimum Euclidean distance between the first, second, and third transition points, and define the ratio as the spatial dispersion index. Calculate the ratio of the area of ​​the spatial base surface to the square of the perimeter of the line connecting the three non-collinear transition points. Compare this ratio with the reciprocal of the spatial dispersion index. When the ratio is less than the reciprocal of the spatial dispersion index, construct a triangular pyramidal shell with the point offset by a preset distance from the geometric constraint origin along the normal of the spatial base surface as the apex and the plane containing the three non-collinear transition points as the base. When the ratio of the area of ​​the spatial base surface to the square of its perimeter is greater than or equal to the reciprocal of the spatial dispersion index, a hyperbolic guiding cylinder is generated by sweeping along the axis with the straight line passing through the origin of the geometric constraint and perpendicular to the spatial base surface as the axis, and the spatial broken line formed by connecting the first transition point, the second transition point and the third transition point in sequence as the generatrix.

4. The method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion according to claim 3, characterized in that, Projecting a triangular pyramidal shell or hyperbolic guiding cylinder onto a spherical coordinate system and performing normal cutting yields a nonlinear thickness slice, including: Project the triangular pyramidal shell or hyperbolic guide cylinder onto a spherical coordinate system with the origin of the geometric constraint as the origin, and take three sampling points at logarithmic intervals in the radial distance; Based on the sampling points, a series of normal cutting planes are constructed. For the triangular pyramid shell, the triangular pyramid shell is decomposed into several triangular facets. The intersection line between each normal cutting plane and each triangular facet is calculated in turn. That is, by determining that the three vertices of the triangular facet are located on the positive side, negative side or plane of the cutting plane, the intersecting edge is determined, and the coordinates of the intersection point are calculated. All intersection points are connected to form the cross-sectional profile of the corresponding cutting plane and the triangular pyramid shell. For a hyperbolic guide cylinder, the hyperbolic guide cylinder is parameterized as a sweeping surface of a hyperbola along the axis. The cross-sectional boundary is obtained by solving the intersection of the cutting plane and the hyperbolic parametric curve. The spatial region enclosed by the cross-sectional profiles obtained by all normal cutting planes is defined as a nonlinear thickness slice, resulting in three nonlinear thickness slices with logarithmic changes in thickness along the polar radius direction.

5. The method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion according to claim 4, characterized in that, Applying a spatial warping transform centered on the spatial guiding pole to each nonlinear thickness slice yields warped slices, including: For each nonlinear thickness slice, a spatial warping transformation is applied. The kernel function of the spatial warping transformation is a Gaussian radial basis function centered at the spatial guiding pole. In the spatial warping transformation, each point on the current nonlinear thickness slice generates a nonlinear radial displacement based on its Euclidean distance from the spatial guiding pole. The magnitude of the radial displacement follows a Gaussian distribution, such that the displacement is larger for points closer to the spatial guiding pole and smaller for points farther away. Each nonlinear thickness slice after the spatial warping transformation is denoted as a warped slice, resulting in three warped slices.

6. The method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion according to claim 5, characterized in that, The maximum eigendirection of the second-order moment of inertia tensor of each warped slice is calculated as the principal inertial axis. The warped slice is divided into sub-bands along the principal inertial axis. The local centroids of each sub-band are extracted and a minimum enclosing ellipsoid is constructed. The center of the ellipsoid is used as the equivalent centroid to obtain the equivalent centroid sequence, including: For each warped slice, the mean of the coordinates of all points on the corresponding warped slice is calculated as the centroid. For each point, the offset vector relative to the centroid is calculated. Based on the offset vector, the six independent components of the second-order moment of inertia tensor are determined, forming a third-order symmetric tensor. The eigenvalues ​​of the third-order symmetric tensor are solved, and the eigendirection corresponding to the largest eigenvalue is taken. The corresponding eigendirection is defined as the principal inertial axis. Divide the current warped slice into six sub-bands along the principal inertial axis, calculate the local centroid coordinates of all points in each sub-band, and obtain six local centroid points; construct a minimum enclosing ellipsoid with the six local centroid points, calculate the center coordinates of the minimum enclosing ellipsoid, and use the center coordinates as the equivalent centroid of the current warped slice; traverse all three warped slices to obtain an equivalent centroid sequence composed of three equivalent centroids.

7. The method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion according to claim 6, characterized in that, The path distortion factor is obtained by calculating the ratio of the total length of the line connecting adjacent equivalent centroids to the straight-line distance between the first and last equivalent centroids, and then applying hyperbolic cotangent mapping. This factor includes: Connect two adjacent equivalent centroids in the equivalent centroid sequence topologically to obtain two spatial lines; calculate the total length of the two spatial lines and the straight-line distance between the first and last equivalent centroids in the equivalent centroid sequence; calculate the ratio of the total length of the spatial lines to the straight-line distance to obtain the initial ratio; apply hyperbolic cotangent mapping to the initial ratio and use the mapped result as the path distortion factor.

8. The method for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion according to claim 7, characterized in that, Step 5 includes: The lightweight AI algorithm engine embedded in the micro AI intelligent body of the probe performs sliding window integration on the original induced electromotive force signal to obtain the energy decay curve. The time corresponding to the peak position in the energy decay curve is extracted as the arrival time of the steel bar reflected echo. At the same time, combined with the real-time position coordinates of the probe in the displacement and attitude data, the initial protective layer thickness value is calculated using the pre-calibrated electromagnetic induction signal amplitude-distance relationship curve. The initial protective layer thickness value is multiplied and combined with the path distortion factor to obtain the corrected protective layer thickness prediction value. The corrected protective layer thickness prediction value is compared with the preset qualified threshold. If it is greater than or equal to the qualified threshold, it is judged as qualified; otherwise, it is judged as unqualified. The corrected protective layer thickness prediction value and the qualified judgment result are output through the host display screen.

9. A system for detecting, analyzing, and predicting the concrete cover of reinforcing bars based on multimodal data fusion, wherein the system implements the method as described in any one of claims 1 to 8, characterized in that, include: The spatial datum construction module is used to acquire the multimodal raw data stream, extract three non-collinear transition points from the multimodal raw data stream, and map them to three-dimensional space to construct the spatial datum. The selection module is used to calculate the centroid and circumcenter of the spatial datum, with the centroid as the origin of the geometric constraint and the circumcenter as the spatial guiding pole. Calculate the spatial dispersion index between three non-collinear transition points, and construct a triangular pyramidal shell or hyperbolic guiding cylinder by combining the spatial distribution characteristics of the spatial datum. The warped slice acquisition module is used to project a triangular pyramidal shell or hyperbolic guiding cylinder onto a spherical coordinate system for normal cutting to obtain a nonlinear thickness slice; and to apply a spatial warping transformation centered on the spatial guiding pole to each nonlinear thickness slice to obtain a warped slice. The path distortion factor calculation module is used to calculate the maximum characteristic direction of the second-order moment of inertia tensor of each warped slice as the principal inertial axis. The warped slice is divided into sub-bands along the principal inertial axis, the local centroid of each sub-band is extracted and the minimum enclosing ellipsoid is constructed. The center of the ellipsoid is used as the equivalent centroid to obtain the equivalent centroid sequence. The ratio of the total length of the line connecting adjacent equivalent centroids to the straight-line distance between the first and last equivalent centroids is calculated and the path distortion factor is obtained through hyperbolic cotangent mapping. The result acquisition module is used to calculate the initial protective layer thickness value, correct the initial protective layer thickness value with the path distortion factor, and obtain the final protective layer thickness prediction value and the qualification judgment result.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 9.

Images