High-precision three-dimensional imaging method and system based on continuous magnetic flux leakage data

Abstract

The invention relates to the technical field of defect detection imaging, and discloses a high-precision three-dimensional imaging method and system based on continuous magnetic flux leakage data. Collecting surface magnetic flux leakage signals of a measured target to form an initial signal set, and extracting feature vectors by using a convolutional neural network model; separating three-dimensional space data, and obtaining initial three-dimensional position coordinates of the defect through an iterative mapping algorithm; mining spatial geometric features, determining potential geometric boundaries, and fusing data to construct a refined position model; extracting a detection subset, correcting a shape error, constructing a space reconstruction grid, filling holes, and obtaining a complete three-dimensional shape representation; and through precision calibration and signal matching verification, a high-precision three-dimensional image is output. According to the method, the defect positioning and shape reconstruction precision can be effectively improved, the imaging integrity and accuracy are guaranteed, the problems of fuzzy positioning and shape distortion of traditional magnetic flux leakage imaging are solved, and the defect detection requirements of a complex detected target are met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of defect detection imaging, and in particular to a high-precision three-dimensional imaging method and system based on continuous magnetic flux leakage data. BACKGROUND

[0002] Industrial detection and non-destructive evaluation is a core technical field to ensure the stable operation of the industrial system, maintain production safety and prolong the service life of equipment, and is widely used in the maintenance of key infrastructure such as pipelines and bridges. Among them, the accurate identification, positioning and shape analysis of internal defects of equipment or materials are the core demands of this field, which directly determines the scientific nature of subsequent maintenance decisions and plays an irreplaceable role in ensuring the long-term safe service of key infrastructure.

[0003] The current mainstream defect detection and imaging technology has obvious limitations when facing complex detection environments and multi-dimensional spatial analysis needs. This kind of technology often has difficulty in fully capturing deep defect information, especially in the signal three-dimensional analysis and spatial feature reconstruction link, often resulting in information loss due to incomplete signal collection and insufficient analysis accuracy, and further causing detection result deviation, which cannot truly reflect the actual shape, distribution range and spatial position of the defect.

[0004] The deeper technical challenges are concentrated in the double difficulties of three-dimensional defect restoration: on the one hand, the surface magnetic flux leakage signal and other collected data are scattered and incomplete, resulting in fuzzy defect spatial positioning and difficulty in accurately locking its three-dimensional coordinates; on the other hand, the positioning ambiguity and signal loss are superimposed, further exacerbating the construction error of the defect geometric shape and boundary range, making the three-dimensional shape restoration significantly deviate from the actual situation. For example, in long-distance pipeline detection, existing technologies can only roughly judge the defect area, but cannot accurately obtain key information such as its depth and specific contour, making it difficult for maintenance personnel to assess the degree of defect damage, and therefore there is an urgent need to break through this technical bottleneck. SUMMARY

[0005] The present application provides a high-precision three-dimensional imaging method and system based on continuous magnetic flux leakage data to effectively improve the defect positioning and shape reconstruction accuracy, ensure the integrity and accuracy of imaging, solve the problems of traditional magnetic flux leakage imaging positioning ambiguity and shape distortion, and adapt to the defect detection needs of complex measured targets.

[0006] In a first aspect, to solve the above technical problems, the present application provides a high-precision three-dimensional imaging method based on continuous magnetic flux leakage data, comprising: acquiring the surface magnetic flux leakage signal of the measured target through a magnetic flux leakage signal collection device to form an initial signal set, and extracting a feature vector of the initial signal set using a preset convolutional neural network model; Three-dimensional spatial data is separated from the initial signal set corresponding to the feature vector, and the preliminary three-dimensional position coordinates of the corresponding defects are calculated by a preset iterative mapping algorithm for the positioning ambiguity areas in the three-dimensional spatial data. Based on the preliminary three-dimensional position coordinates, the spatial geometric features of the defect are mined and the potential geometric boundary of the defect is determined. When the deviation between the potential geometric boundary and the surface leakage magnetic signal is less than a preset deviation threshold, the boundary data corresponding to the potential geometric boundary is fused to construct a refined position model. Based on the refined position model, a detection subset of defects is extracted. When the shape error in the detection subset exceeds a preset allowable range, a preset correction function is applied to correct the detection subset to obtain the corrected shape parameters. Based on the corrected shape parameters, a spatial reconstruction mesh of the defect is constructed, and missing information is filled into the empty areas in the spatial reconstruction mesh to obtain a complete three-dimensional shape representation of the defect. Based on the complete three-dimensional shape representation, three-dimensional accuracy calibration and leakage magnetic signal matching verification are performed. After the verification is qualified, a high-precision three-dimensional image of the defect is output. The detection subset includes local spatial coordinate data, local geometric contour data, local magnetic flux leakage signal characteristic data, and local morphological deviation data of the defect.

[0007] In one optional implementation, the step of acquiring the surface magnetic flux leakage signal of the target under test through a magnetic flux leakage signal acquisition device to form an initial signal set, and extracting the feature vector of the initial signal set using a preset convolutional neural network model, includes: The magnetic flux leakage signal acquisition device is attached to the surface of the target under test to a preset accuracy, and the magnetic flux leakage signal of the entire surface of the target under test is acquired at a preset frequency to obtain the original magnetic flux leakage signal. The original magnetic flux leakage signal is processed by removing invalid signals caused by interference and noise, and then organized according to the acquisition time sequence to form an initial signal set. The initial signal set is input into a preset convolutional neural network model, and the leakage magnetic signal features are extracted and compressed layer by layer through convolutional layers and pooling layers; The leakage magnetic signal features are fused through the fully connected layer of the convolutional neural network model to output a feature vector representing the correlation between the leakage magnetic signal and the defect.

[0008] In one optional implementation, the step of separating three-dimensional spatial data from the initial signal set corresponding to the feature vector, and calculating the preliminary three-dimensional position coordinates of the corresponding defect using a preset iterative mapping algorithm for the positioning ambiguity regions existing in the three-dimensional spatial data, includes: By associating the feature vectors with the initial signal set, three-dimensional spatial data related to defect localization are extracted from the initial signal set; The three-dimensional spatial data is scanned across the entire domain to identify corresponding ambiguous and normal regions, and abnormal data points and data distribution characteristics within the ambiguous regions are marked. The marked ambiguous positioning area and the normal area are input into a preset iterative mapping algorithm, and calibration is performed through iterative mapping to obtain calibrated three-dimensional spatial data; Based on the calibrated three-dimensional spatial data, the preliminary three-dimensional location coordinates of the defect are calculated.

[0009] In one optional implementation, the step of mining the spatial geometric features of the defect based on the preliminary three-dimensional position coordinates and determining the potential geometric boundary of the defect, and when the deviation between the potential geometric boundary and the surface magnetic flux leakage signal is less than a preset deviation threshold, fusing the boundary data corresponding to the potential geometric boundary to construct a refined position model, includes: Based on the preliminary three-dimensional position coordinates, the leakage magnetic signal features of the corresponding area are extracted, and the spatial geometric features of the defects are mined based on the leakage magnetic signal features. Based on the spatial geometric features, the potential geometric boundaries of the defects are determined, and the boundary data corresponding to the potential geometric boundaries are extracted. Calculate the signal deviation between the potential geometric boundary and the surface leakage magnetic field signal, and compare the signal deviation with a preset deviation threshold; When the deviation is less than a preset deviation threshold, the boundary data is fused to construct a refined position model.

[0010] In one optional implementation, the step of extracting a detection subset of defects based on the refined position model, and when the shape error in the detection subset exceeds a preset allowable range, applying a preset correction function to correct the detection subset to obtain corrected shape parameters, includes: Extract a subset of detectors related to defect morphology from the refined location model; The probe subset is subjected to morphological analysis to detect existing shape errors and calculate shape error values; wherein, the shape errors include local distortion deviation of defect contour, local error of defect size, local offset deviation of defect boundary and local distortion deviation of defect morphology. The shape error value is compared with a preset allowable range; When the shape error value exceeds the preset allowable range, a preset correction function is invoked to correct the shape error and obtain the corrected shape parameters.

[0011] In one optional implementation, the step of constructing a spatial reconstruction mesh of the defect based on the corrected shape parameters, and filling missing information in the void regions of the spatial reconstruction mesh to obtain a complete three-dimensional shape representation of the defect, includes: Based on the corrected shape parameters, set the mesh construction parameters and construct a spatial reconstruction mesh for the defects; Perform global detection on the spatial reconstruction mesh to identify and mark the range of void regions; Extract the magnetic flux leakage signal and geometric feature data corresponding to the void region, and fill in the missing information of the void region; After filling in the missing information, a complete three-dimensional shape representation of the defect is obtained.

[0012] In one optional implementation, the step of performing three-dimensional accuracy calibration and magnetic flux leakage signal matching verification based on the complete three-dimensional shape representation, and outputting a high-precision three-dimensional image of the defect after passing the verification, includes: The complete three-dimensional shape representation of the defect is invoked, and a three-dimensional accuracy calibration operation is performed based on the preset calibration standard. During the calibration process, the spatial offset deviation is corrected simultaneously to obtain the calibrated three-dimensional shape data. Extract the calibrated three-dimensional shape data, select the leakage magnetic signal segment corresponding to the initial signal set, perform matching verification between signal features and three-dimensional shape features, and output a high-precision three-dimensional image of the defect.

[0013] Secondly, the present invention also provides a high-precision three-dimensional imaging system based on continuous magnetic flux leakage data, comprising: Signal feature extraction module: Acquires the surface magnetic flux leakage signal of the target under test through a magnetic flux leakage signal acquisition device to form an initial signal set, and extracts the feature vector of the initial signal set using a preset convolutional neural network model; Three-dimensional preliminary positioning module: Separates three-dimensional spatial data from the initial signal set corresponding to the feature vector, and calculates the preliminary three-dimensional position coordinates of the corresponding defect by using a preset iterative mapping algorithm for the positioning ambiguity area in the three-dimensional spatial data; Refined position modeling module: Based on the preliminary three-dimensional position coordinates, the spatial geometric features of the defect are mined and the potential geometric boundaries of the defect are determined. When the deviation between the potential geometric boundary and the surface leakage magnetic signal is less than a preset deviation threshold, the boundary data corresponding to the potential geometric boundary is fused to construct a refined position model. Shape error correction module: Based on the refined position model, extract the detection subset of defects. When the shape error in the detection subset exceeds the preset allowable range, apply the preset correction function to correct the detection subset and obtain the corrected shape parameters. Mesh Reconstruction and Filling Module: Based on the corrected shape parameters, a spatial reconstruction mesh of the defect is constructed, and missing information is filled into the void areas in the spatial reconstruction mesh to obtain a complete three-dimensional shape representation of the defect; 3D calibration imaging module: Based on the complete 3D shape representation, it performs 3D accuracy calibration and leakage magnetic signal matching verification. After the verification is qualified, it outputs a high-precision 3D image of the defect. The detection subset includes local spatial coordinate data, local geometric contour data, local magnetic flux leakage signal characteristic data, and local morphological deviation data of the defect.

[0014] In one optional implementation, the three-dimensional preliminary positioning module includes: Data association and separation unit: associates the correspondence between the feature vector and the initial signal set, and separates the three-dimensional spatial data related to defect location from the initial signal set; Full-domain scanning and marking unit: performs full-domain scanning on the three-dimensional spatial data, identifies corresponding ambiguous and normal regions, and marks abnormal data points and data distribution characteristics within the ambiguous regions; Iterative mapping calibration unit: The marked positioning fuzzy area and the normal area are input into a preset iterative mapping algorithm, and calibration is performed through iterative mapping to obtain calibrated three-dimensional spatial data; Three-dimensional coordinate calculation unit: Based on the calibrated three-dimensional spatial data, the preliminary three-dimensional position coordinates of the defect are calculated.

[0015] In one optional implementation, the shape error correction module includes: Detector subset extraction unit: Extracts a detector subset related to the defect morphology from the refined location model; Shape error detection unit: performs shape analysis on the detection subset, detects existing shape errors and calculates shape error values; wherein, the shape errors include local distortion deviation of defect contour, local error of defect size, local offset deviation of defect boundary and local distortion deviation of defect shape; Error range comparison unit: compares the shape error value with a preset allowable range; Shape parameter correction unit: When the shape error value exceeds the preset allowable range, a preset correction function is called to correct the shape error and obtain the corrected shape parameters.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) Achieve precise defect localization. By extracting the feature vector of the magnetic flux leakage signal through a convolutional neural network model and combining iterative mapping algorithm to calibrate the ambiguous localization area in the three-dimensional spatial data, the correlation between the defect and the spatial data is accurately quantified, effectively solving the problems of ambiguous localization and large coordinate deviation in traditional technology, and providing a reliable position benchmark for subsequent morphological reconstruction.

[0017] (2) Improve the integrity of morphological reconstruction. Based on the refined position model, extract the probe subset containing multi-dimensional data, correct the shape error through the correction function, and then fill the holes in the spatial reconstruction mesh to completely restore the three-dimensional morphology of the defect, avoid shape distortion caused by signal loss, and ensure the consistency between the imaging and the actual defect.

[0018] (3) Enhance the controllability of imaging accuracy. Construct a closed-loop technology process of "preliminary positioning - refined modeling - error correction - accuracy calibration". Through multiple rounds of data verification and signal matching, the imaging accuracy is controlled at each level, significantly reducing the interference of complex detection environment on the results and meeting the core requirements of industrial detection for high-precision imaging.

[0019] (4) Enhance the flexibility of scene adaptation. Relying on feature vector extraction and semantic association analysis, it can adapt to the surface characteristics and defect types of different test targets without adjusting the core algorithm for specific scenes. This effectively breaks through the limitations of insufficient adaptability of traditional technologies and broadens the application scope of industrial non-destructive testing.

[0020] (5) Ensure the scientific nature of detection decisions. The output high-precision three-dimensional images fully contain key information such as defect location, geometry, and boundary range, providing comprehensive data support for equipment maintenance and risk assessment, avoiding decision-making biases caused by incomplete detection information, and helping to improve the operational safety of the industrial system. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a high-precision three-dimensional imaging method based on continuous magnetic flux leakage data provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a high-precision three-dimensional imaging system based on continuous magnetic flux leakage data provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Reference Figure 1 This invention provides a high-precision three-dimensional imaging method based on continuous magnetic flux leakage data, comprising the following steps: S11, the surface magnetic flux leakage signal of the target under test is acquired through the magnetic flux leakage signal acquisition device to form an initial signal set, and the feature vector of the initial signal set is extracted using a preset convolutional neural network model; S12, separate three-dimensional spatial data from the initial signal set corresponding to the feature vector, and calculate the preliminary three-dimensional position coordinates of the corresponding defect by using a preset iterative mapping algorithm for the positioning ambiguity area in the three-dimensional spatial data. S13, Based on the preliminary three-dimensional position coordinates, the spatial geometric features of the defect are mined and the potential geometric boundary of the defect is determined. When the deviation between the potential geometric boundary and the surface leakage magnetic signal is less than a preset deviation threshold, the boundary data corresponding to the potential geometric boundary is fused to construct a refined position model. S14. Based on the refined position model, extract the detection subset of defects. When the shape error in the detection subset exceeds the preset allowable range, apply the preset correction function to correct the detection subset and obtain the corrected shape parameters. S15, Based on the corrected shape parameters, construct a spatial reconstruction mesh for the defect, and fill the missing information in the void areas of the spatial reconstruction mesh to obtain a complete three-dimensional shape representation of the defect; S16. Based on the complete three-dimensional shape representation, perform three-dimensional accuracy calibration and leakage magnetic signal matching verification. After the verification is qualified, output a high-precision three-dimensional image of the defect.

[0024] In step S11, the surface magnetic flux leakage signal of the target under test is acquired by the magnetic flux leakage signal acquisition device to form an initial signal set, and the feature vector of the initial signal set is extracted by a preset convolutional neural network model.

[0025] In one implementation, the core of this step is to accurately acquire the magnetic flux leakage signal and efficiently extract the defect correlation features, providing a high-quality data foundation for subsequent three-dimensional imaging. All terminology and technical details are consistent with the actual industrial inspection scenario. The magnetic flux leakage signal acquisition device employs a high-precision Hall sensor array (8-channel parallel acquisition), possessing anti-electromagnetic interference capabilities. It can directly attach to the surface of the target to acquire magnetic field intensity change signals, with a magnetic field measurement range of ±20mT and a measurement accuracy of ±0.01mT. This is the core hardware for capturing defect magnetic flux leakage characteristics. The preset attachment accuracy refers to the contact gap error threshold between the sensor array and the target surface, set to ±0.1mm in this embodiment. This accuracy avoids signal attenuation caused by excessive gaps and prevents equipment wear or signal distortion caused by excessively tight contact. The preset acquisition frequency is set to 1kHz to ensure signal continuity and feature integrity, i.e., acquiring 1000 sets of full-domain magnetic flux leakage signals per second. This can completely cover the details of magnetic field changes in the defect area and avoid the loss of high-frequency features. The convolutional neural network model uses 3 convolutional layers (Conv1d) + 2 max pooling layers (MaxPool1d) + 2 The lightweight structure of the fully connected layer (Linear) is specifically adapted for one-dimensional temporal feature extraction of leakage magnetic signals, balancing computational efficiency and feature extraction accuracy. The feature vector is a high-dimensional quantized data characterizing the correlation between leakage magnetic signals and defects. In this embodiment, the output dimension is 128-dimensional, and each dimension corresponds to a specific defect-related feature (such as magnetic field gradient change, signal amplitude distribution, continuous fluctuation frequency, etc.). The vector value range is [0,1], and the closer the value is to 1, the stronger the correlation between the feature and the defect.

[0026] In the specific implementation process, the raw magnetic flux leakage signal is first collected: the magnetic flux leakage signal collection device is attached to the surface of a DN500mm industrial carbon steel pipe using a magnetic fixing device. The attachment gap is calibrated using a laser rangefinder to ensure that the attachment error of all sensor units is ≤ ±0.1mm (meeting the preset attachment accuracy). The collection system is started and continuously scans the entire surface of the pipe axially for 10m and circumferentially for 360° at a preset frequency of 1kHz for 60 seconds, acquiring a total of 60×1000=60000 sets of raw magnetic flux leakage signals. Each set of signals contains magnetic field strength data (unit: mT) from 8 channels. Then, invalid signals are removed and the initial signal set is constructed: the raw magnetic flux leakage signal contains invalid signals such as electromagnetic interference (e.g., power frequency interference in the industrial environment) and mechanical vibration noise (e.g., friction noise when the collection device moves). Wavelet threshold denoising algorithm is used for processing, with the wavelet basis set to db4 and the decomposition level set to 5. The threshold calculation formula is as follows (Formula 1): Where λ is the wavelet threshold, σ is the noise standard deviation (σ=0.02 is obtained by calculating the standard deviation of the high-frequency coefficients of the 5th layer after wavelet decomposition of the original signal), and N is the signal length (N=60000 in this step). Substituting these values, we get λ≈0.02×10.8≈0.216. The high-frequency coefficients after wavelet decomposition are truncated using this threshold to retain the effective signal components. The signal is then reconstructed by inverse wavelet transform to obtain the denoised pure magnetic flux leakage signal. The pure magnetic flux leakage signal is then organized according to the three-dimensional time sequence dimension of "axial position - circumferential angle - acquisition time" to form an initial signal set of 10m (axial) × 360° (circumferential) × 1000 (acquisition frequency). The data format is a matrix (dimension: 1000×360×1000).Finally, convolutional neural network feature extraction and feature vector output are performed: the initial signal set is split according to the channel dimension and input into the preset convolutional neural network model, sequentially passing through the 1st convolutional layer (Conv1d-1: 32 kernels, 3×3 size, stride 1, "same" padding, ReLU activation, output dimension 1000×360×32), the 1st pooling layer (MaxPool1d-1: 2×2 pooling windows, stride 2, output dimension 500×180×32), the 2nd convolutional layer (Conv1d-2: 64 kernels, 3×3 size, stride 1, "same" padding, ReLU activation, output dimension 500×180×64), the 2nd pooling layer (MaxPool1d-2: 2×2 pooling windows, stride 2, output dimension 250×90×64), the 3rd ...4th pooling layer (MaxPool1d-2: 2×2 pooling windows, stride 2, output dimension 250×90×64), the 4th pooling layer (Conv1d-2: 64 kernels, 3×3 size, stride 1, "same" padding, ReLU activation, output dimension 500×180×64), the 4th pooling layer (Max The convolutional layer (Conv1d-3: 128 kernels, 3×3 size, stride 1, "same" padding, ReLU activation, output dimension 250×90×128) flattens the feature map output by Conv1d-3, transforming it into a one-dimensional feature vector (length = 250×90×128 = 2880000). This vector is then input to the fully connected layer for feature fusion. The first fully connected layer (Linear-1: input dimension 2880000, output dimension 1024, ReLU activation) performs linear transformation and non-linear mapping on the flattened features. The second fully connected layer (Linear-2: input dimension 1024, output dimension 128, Sigmoid activation) compresses the features to the [0,1] interval, fusing them to generate a 128-dimensional feature vector. The first 30 elements of this feature vector... The high-weighted features (weight ≥ 0.7) correspond to core related information such as defect existence, approximate defect size, and defect location trend, providing key basis for subsequent 3D spatial data separation and defect localization. They are also the core data foundation for subsequent image generation. Each feature in the vector corresponds to key parameters required for 3D imaging (such as magnetic field strength distribution characteristics determining the brightness gradient of the image, and signal continuous fluctuation characteristics determining the contour smoothness of the image). By accurately extracting the feature vector, it can be ensured that the magnetic field correlation features of the defect can be accurately restored in the subsequent image generation process, laying the foundation for the accuracy and integrity of the imaging.

[0027] In step S12, three-dimensional spatial data is separated from the initial signal set corresponding to the feature vector, and a preset iterative mapping algorithm is used to calculate the preliminary three-dimensional position coordinates of the corresponding defect for the positioning ambiguity region in the three-dimensional spatial data.

[0028] In one implementation, the core of this step is to accurately separate the three-dimensional spatial data associated with defects from the initial signal set, and to calibrate and locate the ambiguous areas through an iterative mapping algorithm, so as to provide a precise spatial location reference for subsequent defect morphology reconstruction and image generation. All terminology and technical details are in line with the actual application scenario of industrial pipeline inspection. Among them, the three-dimensional spatial data is the three-dimensional quantitative data characterizing the spatial distribution of the leakage magnetic signal on the surface of the target under test. It includes three dimensions: axial (x-axis, unit: m), circumferential (y-axis, unit: °), and radial (z-axis, unit: mm), which are directly related to the spatial location information of the defect. The fuzzy positioning region refers to the area in the three-dimensional spatial data where the defect location has low identification due to signal interference, acquisition blind spots, etc. The data points in this region are discretely distributed, and it is impossible to directly lock the accurate coordinates of the defect. The iterative mapping algorithm adopts an improved K-means iterative mapping algorithm, which calculates the cluster center of the data points through multiple iterations to achieve calibration and optimization of the fuzzy region data. Its core advantages are fast convergence speed and strong anti-interference ability. The preliminary three-dimensional position coordinates refer to the preliminary quantitative coordinates of the defect center in three-dimensional space obtained after algorithm calibration. The coordinate accuracy needs to meet ±0.05mm, which provides a basic position reference for subsequent refined modeling.

[0029] In the specific implementation process, the three-dimensional spatial data is first separated: based on the correlation mapping relationship between the 128-dimensional feature vector output in step S11 and the initial signal set, the highly correlated dimensions (30 dimensions in total) with a weight ≥ 0.7 in the feature vector are extracted. These dimensions directly correspond to the magnetic field distribution characteristics of the defect. Using these 30-dimensional features as the screening criteria, the three-dimensional spatial data related to defect location are separated from the initial signal set of 1000×360×1000 dimensions. The data format is (x - axial position, y - circumferential angle, z - defect depth correlation value corresponding to magnetic field strength). The data is then normalized to a uniform dimension (500×360×500) by linear interpolation to ensure the continuity and consistency of the data. Subsequently, ambiguous region identification was performed: a sliding window method (window size 5×5×5) was used to scan the entire area of ​​the normalized 3D spatial data. A data dispersion threshold of 0.3 (dispersion = standard deviation / mean) was set. When the data dispersion within the window was ≥0.3, the area was determined to be an ambiguous region. Abnormal data points (points where the magnetic field strength deviates from twice the mean of the region) and data distribution characteristics (continuous clustered distribution / discrete point distribution) were marked within the region. A total of 3... The locations of the ambiguous regions are identified as follows: Region 1 (x: 2.3-2.5m, y: 90-100°, z: 2.5-3.0mm), Region 2 (x: 5.1-5.3m, y: 220-230°, z: 1.2-1.8mm), and Region 3 (x: 7.8-8.0m, y: 310-320°, z: 0.6-1.0mm). The proportions of abnormal data points in each region are 35%, 28%, and 22%, respectively, all exhibiting a continuous clustered distribution. Next, data calibration is performed: the labeled ambiguous region data and normal region data are input into a preset improved K-means iterative mapping algorithm. The maximum number of iterations is set to 50, and the convergence threshold is 0.001 (meaning the cluster center coordinate deviation between two adjacent iterations is ≤0.001mm; the algorithm is considered converged when this condition is met for three consecutive iterations). A spatial distance weighting factor is introduced during the algorithm calibration process, calculated using the following formula (Formula 2): Among them, w ij x represents the spatial distance weight between data point i and cluster center j. i y i z i Let x be the three-dimensional coordinates of data point i. j y j z jLet $k$ be the 3D coordinates of cluster center $j$, and $k$ be the radial weight coefficient (valued at 5.0, as radial data is more critical for defect depth localization). This formula assigns differentiated weights to data points at different spatial locations, improving calibration accuracy. After 32 iterations, the algorithm reaches the convergence threshold, yielding calibrated 3D spatial data. The data dispersion within each fuzzy region is reduced to below 0.15. Finally, preliminary 3D position coordinates are calculated: based on the calibrated 3D spatial data, the cluster center coordinates of the calibrated data points within each fuzzy region are extracted. These cluster centers are the preliminary 3D position coordinates of the defects. The calculation results are: Defect 1 (2.40m, 95.0°, 2.78mm), Defect 2 (5.20m, 225.0°, 1.52mm), and Defect 3 (7.90m, 315.0°, 0.83mm), with a coordinate accuracy of ±0.03mm, meeting the positioning accuracy requirements of subsequent steps.

[0030] The preliminary 3D position coordinates output in this step are the core spatial reference for image generation. The accuracy of the coordinates directly determines the accuracy of the defect location restoration in subsequent 3D imaging. By calibrating the blurred areas through an iterative mapping algorithm, the positioning deviation problem caused by signal interference in traditional technology is effectively solved, laying a solid positional foundation for generating high-fidelity 3D defect images.

[0031] In step S13, based on the preliminary three-dimensional position coordinates, the spatial geometric features of the defect are mined and the potential geometric boundary of the defect is determined. When the deviation between the potential geometric boundary and the surface leakage magnetic signal is less than a preset deviation threshold, the boundary data corresponding to the potential geometric boundary is fused to construct a refined position model.

[0032] In one implementation, the core of this step is based on the preliminary three-dimensional position coordinates, accurately mining the spatial geometric features of the defect and determining the potential geometric boundaries. A high-precision refined position model is constructed through deviation verification and boundary data fusion, providing a core modeling foundation for subsequent image generation that combines positional accuracy and morphological integrity. All terminology and technical details are in line with the actual scenario of industrial pipeline defect detection. Among them, spatial geometric features are the core attribute set characterizing the three-dimensional morphology of defects, including the shape type of defects (such as elliptical corrosion, linear cracks, point defects), size parameters (major axis / minor axis length, depth, width), contour curvature and other quantitative indicators, which directly determine the accuracy of the three-dimensional morphology reconstruction of defects; potential geometric boundaries refer to the virtual contours that may characterize the actual boundaries of defects, derived from leakage magnetic signal features and spatial geometric features, usually presented in the form of a set of three-dimensional coordinate points or mathematical equations, which is a preliminary quantitative description of the defect morphology; the preset deviation threshold of 0.05mT is set at twice the existing leakage magnetic signal noise level (≤0.02mT), which avoids misjudgment and ensures the accuracy of boundary recognition; the refined position model is a high-precision defect model formed by integrating reliable potential geometric boundary data. It not only contains accurate three-dimensional position coordinates, but also integrates key information such as defect boundaries, size, and shape. The model position accuracy reaches ±0.01mm, and the morphology reconstruction degree is better than 95%.

[0033] In the specific implementation process, spatial geometric feature mining is first performed: using the preliminary three-dimensional position coordinates of the three defects obtained in step S12 (defect 1: 2.40m, 95.0°, 2.78mm; defect 2: 5.20m, 225.0°, 1.52mm; defect 3: 7.90m, 315.0°, 0.83mm) as a benchmark, a spherical neighborhood screening method with a radius of 5mm is used to extract the leakage magnetic signal features of the corresponding regions, including signal amplitude distribution, magnetic field gradient change, and signal duration. Among them, the leakage magnetic signal amplitude of the region corresponding to defect 1 is 8.2-9.1mT, showing a symmetrical single-peak distribution, with a gentle change in magnetic field gradient, and a signal duration of 8mm along the axial direction and 6mm in the circumferential direction. Combined with the preset leakage magnetic signal-defect feature mapping library (containing 1000+ The signal feature template of typical defects in industrial pipelines was used to match and determine that the spatial geometric feature of defect 1 was an elliptical corrosion pit, with preliminary size parameters of 8mm major axis, 6mm minor axis, and 2.78mm depth. The magnetic flux leakage signal amplitude of defect 2 was 4.3-5.1mT, showing a linear continuous distribution, with a signal duration of 12mm circumferentially and 0.3mm axially, and was determined to be a linear crack, with preliminary size parameters of 12mm length, 0.3mm width, and 1.52mm depth. The magnetic flux leakage signal amplitude of defect 3 was 2.1-2.6mT, showing a point-like concentrated distribution, with a signal coverage area diameter of 3mm, and was determined to be point corrosion, with preliminary size parameters of 3mm diameter and 0.83mm depth. Subsequently, the potential geometric boundary was determined and the boundary data was extracted: based on the spatial geometric features of each defect, the potential geometric boundary was constructed using mathematical modeling, where the boundary of the elliptical corrosion pit (defect 1) was represented by the ellipse equation (Formula 3). In the formula, (x0, y0) represents the axial-circumferential coordinates of the defect center (2.40m, 95.0°), a is the major axis radius (4mm), b is the minor axis radius (3mm), and the radial boundary is z=2.78mm; the boundary of the linear crack (defect 2) is characterized by a spatial straight line equation combined with the width range: (Extending circumferentially), the width boundary is x±0.15mm; the boundary of the pitting corrosion (defect 3) is characterized by the spherical equation: In the formula, r is the corrosion radius (1.5 mm). Discrete coordinate points on each potential geometric boundary are extracted as boundary data. 100 boundary points are extracted for defect 1, 80 boundary points for defect 2, and 60 boundary points for defect 3, forming a boundary data set. Next, signal deviation calculation and comparison are performed: the mean square error (MSE) algorithm is used to calculate the signal deviation between the potential geometric boundary and the corresponding surface magnetic flux leakage signal. The calculation formula is as follows (Formula 4): Where n is the number of boundary data points, B iLet B be the amplitude of the actual surface leakage magnetic signal corresponding to the i-th boundary point. i Let ′ be the theoretical leakage magnetic signal amplitude of the i-th boundary point derived from the potential geometric boundary model; substituting the data, we calculate that: the MSE of defect 1 is 0.0009mT² (deviation value 0.03mT), the MSE of defect 2 is 0.0004mT² (deviation value 0.02mT), and the MSE of defect 3 is 0.0016mT² (deviation value 0.04mT), all of which are less than the preset deviation threshold of 0.05mT, indicating that each potential geometric boundary is reliable. Finally, a refined location model was constructed: reliable potential geometric boundary data of each defect were fused with the preliminary 3D location coordinates, and a weighted average method was used to optimize the accuracy of the boundary point coordinates (the boundary point weight was 0.7, and the preliminary coordinate weight was 0.3). At the same time, information such as the size parameters and shape type of the defects was integrated to construct a refined location model of the three defects. In this model, the precise coordinates of defect 1 are (2.402m, 95.01°, 2.778mm), with a major axis of 7.98mm and a minor axis of 5.99mm; the precise coordinates of defect 2 are (5.201m, 225.02°, 1.521mm), with a length of 11.97mm and a width of 0.298mm; and the precise coordinates of defect 3 are (7.903m, 315.01°, 0.829mm), with a diameter of 2.99mm. The positional accuracy and morphological accuracy of the model are both improved by an order of magnitude compared with the preliminary positioning stage, providing core data support for the accurate restoration of defect morphology in subsequent image generation.

[0034] In step S14, based on the refined position model, a detection subset of defects is extracted. When the shape error in the detection subset exceeds a preset allowable range, a preset correction function is applied to correct the detection subset to obtain the corrected shape parameters.

[0035] In one implementation, the core of this step is to extract a defect detection subset based on a refined location model, identify shape errors through morphological analysis, and correct errors using a preset correction function. This provides high-precision shape data for subsequent spatial reconstruction and image generation. All terminology and technical details are tailored to the actual application scenario of industrial pipeline defect detection. The detection subset is a multi-dimensional data set directly related to the defect morphology extracted from the refined location model. Specifically, it includes local spatial coordinate data of the defect (three-dimensional coordinates of the boundary and internal feature points), local geometric contour data (cross-sectional / longitudinal section contour curve parameters), local magnetic flux leakage signal characteristic data (signal amplitude and phase distribution in the defect area), and local morphological deviation data (quantized deviation values ​​between the preliminary contour and the standard template). This is the core data foundation for morphological analysis and error correction. Shape error refers to the deviation between the defect morphology represented by the detection subset and the actual defect morphology. Specifically, it includes local distortion deviations in the defect contour (irregular bulges or depressions in the contour curve), local errors in defect dimensions (major axis / minor axis), and other related data. The deviations of key dimensions such as minor axis, diameter, and depth from their actual values, local offset deviations of defect boundaries (deviations between boundary point coordinates and actual boundaries), and local distortion deviations of defect morphology (inconsistencies between the overall shape and the actual type) are all represented by quantitative values. The preset allowable ranges are determined based on the high-precision requirements of industrial inspection and the measurement capabilities of the equipment, specifically: local contour distortion deviation ±0.02mm, local size error ±0.1mm, local boundary offset deviation ±0.03mm, and local morphological distortion deviation ≤5%. The correction function adopts a polynomial fitting correction function. Through fitting analysis of error data, the corrected shape parameters are output. The core advantage is that it can accurately match the error distribution characteristics of complex defects. The corrected shape parameters are quantitative parameters that are highly consistent with the actual defect morphology after correction by the correction function, including the corrected dimensions, contour curve equation, boundary coordinates, etc., providing an accurate basis for the morphological restoration of image generation.

[0036] In the specific implementation process, the first step is to extract the detection subset: from the refined location model constructed in step S13, detection subsets are extracted for each of the three defects. Taking defect 1 (elliptical corrosion pit) as an example, the extracted local spatial coordinate data includes the three-dimensional coordinates of 1200 feature points on the boundary and inside (x accurate to 0.001m, y accurate to 0.01°, z accurate to 0.001mm); the local geometric contour data includes the axial and circumferential cross-sectional contour curves, expressed as a quadratic function. The characteristics were defined as follows: axial profile curve parameters a=0.003, b=0.012, c=2.778; circumferential profile curve parameters a=0.002, b=0.008, c=5.99; local magnetic flux leakage signal characteristic data consisted of the signal amplitude distribution collected by 8-channel sensors within the defect area, ranging from 8.1 to 9.0 mT, with an average value of 8.55 mT; local morphological deviation data consisted of the deviation values ​​between the preliminary profile and the preset elliptical corrosion pit standard template, with a major axis deviation of 0.15 mm, a minor axis deviation of 0.01 mm, and a depth deviation of 0.005 mm. The detection subsets of defects 2 (linear cracks) and 3 (pitting corrosion) were extracted along the same dimensions, containing coordinates of 1000 and 800 feature points respectively, along with corresponding profile data, signal data, and deviation data. Subsequently, morphological analysis and shape error calculation were performed: The contour comparison method and dimensional measurement method were used to perform morphological analysis on the probe subset. The shape error value was calculated by comparing the contour data of the probe subset with the standard contour data in the refined position model point by point. To accurately quantify the shape error, an error calculation formula was introduced: Where ΔL is the average dimensional error, n is the number of measurement points (n=50 in this embodiment), and L i To detect the size measurement of the subset, L 0i To refine the standard dimensional values ​​in the position model. Calculations show that defect 1 has a maximum local distortion deviation of 0.018mm (within the preset allowable range), a major axis dimensional error ΔL = 0.15mm (exceeding the ±0.1mm allowable range), a minor axis dimensional error of 0.01mm (within the allowable range), a maximum local boundary offset deviation of 0.025mm (within the preset allowable range), and a local shape distortion deviation of 3% (within the allowable range). Defect 2 has a local distortion deviation of 0.01mm, a dimensional error of 0.03mm, a boundary offset deviation of 0.02mm, and a shape distortion deviation of 2%, all within the preset allowable range. Defect 3 also has no shape errors exceeding the preset allowable range, except for defect 1 which has a major axis dimensional error exceeding the limit. Next, shape error correction is performed: For the major axis dimensional error of defect 1, a preset polynomial fitting correction function is called, with the function expression as follows: L corr For the corrected dimensions, L err The values ​​are the original error values, and p0, p1, and p2 are the fitting coefficients (obtained by least squares optimization using training data of 100 sets of standard defects in industrial pipelines (with known actual dimensions and measurement errors)). The values ​​are p0 = 7.98, p1 = -0.85, and p2 = 0.12. The original error values ​​are obtained by training with historical error correction data. 0.85, p2=0.12. The major axis dimension error L of defect 1...err Substituting 0.15mm into the formula, we get L. corr =7.98+( 0.85)×0.15+0.12×(0.15) 2 ≈7.98 - 0.1275 + 0.0027 ≈ 7.855 mm. Simultaneously, based on the corrected major axis dimension, the curve parameters of the local geometric contour data are also corrected. The axial contour curve parameters are updated to a = 0.0028, b = 0.011, and c = 2.778, while the circumferential contour curve parameters remain unchanged. Finally, the corrected shape parameters are output: The corrected shape parameters for the three defects are as follows: Defect 1 (elliptical corrosion pit): major axis 7.85 mm, minor axis 5.99 mm, depth 2.778 mm, axial contour curve... Circumferential profile curve Defect 2 (linear crack): length 11.97mm, width 0.298mm, depth 1.521mm, profile curve is a straight line equation y=0.02x+225.02; Defect 3 (pitting corrosion): diameter 2.99mm, depth 0.829mm, profile curve is a circle equation. After verification, the corrected shape parameters matched the actual defect morphology with a accuracy of 99.2%, providing core data support for high-precision morphological restoration in subsequent spatial mesh reconstruction and image generation, ensuring that the generated 3D image can accurately replicate the actual size and contour features of the defect.

[0037] In step S15, based on the corrected shape parameters, a spatial reconstruction mesh of the defect is constructed, and missing information is filled into the void regions in the spatial reconstruction mesh to obtain a complete three-dimensional shape representation of the defect.

[0038] In one implementation, the core of this step is to construct a defect space reconstruction mesh based on the corrected shape parameters. By identifying void regions and filling missing information, a complete three-dimensional shape representation of the defect is obtained, providing a complete and distortion-free morphological data foundation for the final image generation. All terminology and technical details are consistent with the actual scenario of industrial pipeline defect detection. Among them, the spatial reconstruction mesh is a digital model that discretizes the three-dimensional shape of the defect into a finite number of mesh units using a specific topological structure (such as tetrahedron or hexahedron). The size of the mesh unit directly affects the imaging accuracy. In this embodiment, it is set to 0.01mm×0.01mm×0.01mm, which can accurately restore the details of tiny defects. Hole regions refer to blank areas in the spatial reconstruction mesh that are not covered by mesh units due to signal acquisition blind spots, data loss, etc. They are characterized by discontinuous meshes and gaps. If they are not filled, the image generation will be distorted. Missing information filling refers to supplementing the three-dimensional coordinates and morphological data of the blank areas based on the effective data around the hole regions through interpolation, fitting and other algorithms to ensure the integrity of the mesh. The complete three-dimensional shape representation is a digital representation that is highly consistent with the actual defect shape after the hole filling. It includes full-dimensional data such as mesh model, boundary coordinates, and size parameters, providing core morphological support for image generation.

[0039] In the specific implementation process, the spatial reconstruction mesh is first constructed: based on the corrected shape parameters output in step S14, the core parameters of mesh construction are set, including the mesh unit size of 0.01mm×0.01mm×0.01mm, the mesh topology type of tetrahedral mesh (adapting to the restoration of complex defect morphology), and the mesh density threshold of 1000 units per cubic millimeter. The Delaunay triangulation algorithm was chosen because of its advantages of adaptive defect boundary and low mesh distortion rate. It can accurately match the spatial morphology of different types of defects, such as elliptical, linear, and spherical defects, ensuring the consistency between the reconstructed mesh and the actual defect. This algorithm was used to mesh the defects based on their corrected shape parameters. For example, defect 1 (elliptical corrosion pit) has corrected shape parameters of 7.85mm major axis, 5.99mm minor axis, and 2.778mm depth, resulting in a spatial reconstruction mesh containing 8200 tetrahedral mesh elements, with the mesh coverage perfectly matching the actual defect size. Defect 2 (linear crack) has corrected parameters of 11.97mm length, 0.298mm width, and 1.521mm depth, forming a linear extension mesh containing 5400 mesh elements. Defect 3 (point corrosion) has corrected parameters of 2.99mm diameter and 0.829mm depth, forming a spherical mesh containing 14100 mesh elements. To quantify mesh quality, a formula for calculating mesh distortion rate was introduced. Where η is the mesh distortion rate, V act V represents the actual mesh cell volume. idealThe volume of an ideal tetrahedral mesh element (in this embodiment, it is approximately 1.1785 × 10⁻⁸ mm) is given. 3 The average distortion rate of each defective mesh was calculated to be 0.92, which is within the acceptable range of 0.8-1.2, ensuring that the mesh shape is regular and there is no serious distortion.

[0040] Subsequently, void region identification was performed: a connectivity analysis algorithm was used to perform full-domain detection on the spatially reconstructed mesh, with a mesh cell connectivity threshold set to 0.02 mm (i.e., adjacent mesh cells with a spacing exceeding this value were considered disconnected). The continuity of the mesh was detected by traversing each cell. The detection revealed two void regions in the spatially reconstructed mesh of defect 2 (linear crack): region A (x: 5.20 m, y: 226.5°-227.0°, z: 1.52 mm) and region B (x: 5.21 m, y: 228.0°-228.5°, z: 1.51 mm), with void areas of 0.05 mm² and 0.03 mm², respectively. Both voids were due to data loss caused by the blind spot in the crack tip signal acquisition. The meshes of defects 1 and 3 had no void regions, and their continuity met the standard.

[0041] Next, missing information was filled: For the two void areas of defect 2, effective magnetic flux leakage signals and geometric feature data within a 5mm radius were extracted—the magnetic flux leakage signal amplitude around area A was 4.2-4.5mT, the magnetic field gradient direction was along the crack extension direction (30°), and the geometric features were linear extension and uniform width; the magnetic flux leakage signal amplitude around area B was 4.1-4.3mT, the magnetic field gradient direction was consistent with area A, and the geometric features matched the linear trend of the surrounding mesh. The radial basis function interpolation algorithm was used to fill the missing information. The interpolation function expression is: ,in For radial basis functions (ε=0.01 is the shape parameter), (x i ,y i ,z i ) represents the coordinates of the effective grid nodes surrounding the cavity, λ i is the interpolation coefficient (solved using the least squares method), and n is the number of surrounding valid nodes (n=20 in this embodiment). The interpolation algorithm calculates the 3D coordinates of each missing node within the hole area, generating supplementary mesh cells. After filling, the hole area is completely covered, achieving 100% mesh continuity.

[0042] Finally, a complete 3D shape representation was obtained: After filling in the missing information, the spatial reconstruction mesh of all defects was integrated and optimized, redundant mesh units were removed, and minor misaligned nodes were corrected to form a complete 3D shape representation of each defect. Among them, the complete 3D shape of defect 1 is a regular elliptical corrosion pit, with a total of 8200 mesh units and dimensions of 7.85mm major axis, 5.99mm minor axis, and 2.778mm depth; defect 2 is a continuous linear crack, with a total of 5400+800=6200 mesh units (including 800 filling units), and dimensions of 11.97mm length, 0.298mm width, and 1.521mm depth; defect 3 is a spherical point corrosion, with a total of 3050 mesh units and dimensions of 2.99mm diameter and 0.829mm depth. This complete 3D shape representation contains all-dimensional data of the defect, including its global spatial coordinates, contour shape, and size parameters. It provides high-fidelity, complete morphological data support for subsequent image generation, ensuring that the generated 3D image can completely replicate the actual spatial shape and detailed features of the defect.

[0043] In step S16, based on the complete three-dimensional shape representation, three-dimensional accuracy calibration and leakage magnetic signal matching verification are performed. After the verification is qualified, a high-precision three-dimensional image of the defect is output.

[0044] In one implementation, the core of this step is to perform accuracy calibration and signal matching verification based on the complete three-dimensional shape representation, and finally output high-precision three-dimensional defect imaging. This is the key finishing step in image generation, and all the definitions and technical details are in line with the actual application needs of industrial pipeline defect detection. The preset calibration standard adopts the ISO 10360-10 three-dimensional measurement accuracy standard, and the spatial offset deviation correction threshold of ±0.01mm is set according to the accuracy requirements of the "three-dimensional inspection of industrial parts" category in this standard. Three-dimensional accuracy calibration refers to correcting problems such as spatial position offset and size deviation that may exist in the complete three-dimensional shape representation by comparing with standard reference data, so as to ensure the consistency between the data and the actual defects. Spatial offset deviation refers to the deviation between the coordinate system of the complete three-dimensional shape representation and the actual defect spatial coordinate system, including translation deviation, rotation deviation, etc., which need to be corrected through calibration. The matching verification of signal features and three-dimensional shape features refers to verifying the logical correlation between the calibrated three-dimensional shape data and the original magnetic flux leakage signal features, so as to ensure that the shape restoration does not deviate from the actual signal support. High-precision three-dimensional imaging is a digital image that accurately reflects the position, shape and size of defects after full-process processing. It is the final result of image generation of this invention, with an imaging accuracy of ±0.03mm, which can be directly used for industrial inspection decision-making.

[0045] In the specific implementation process, three-dimensional accuracy calibration is first performed: The complete three-dimensional shape representation data of the three defects obtained in step S15 is retrieved, and a standard calibration target (a high-precision metal marker with known three-dimensional coordinates, coordinate accuracy ±0.001mm) is introduced for comparison and calibration based on the ISO 10360-10 standard. The spatial offset deviation between the complete three-dimensional shape representation and the calibration target is calculated using the least squares method. The formula for calculating the offset deviation is: Where ΔP is the comprehensive spatial offset deviation, and Δx, Δy, and Δz are the axial, circumferential, and radial offset components, respectively. Calculations show that for defect 1, Δx = 0.008 mm, Δy = 0.005 mm, and Δz = 0.003 mm, with a comprehensive offset deviation ΔP ≈ 0.010 mm; for defect 2, Δx = 0.006 mm, Δy = 0.004 mm, and Δz = 0.002 mm, with a comprehensive offset deviation ΔP ≈ 0.007 mm; and for defect 3, Δx = 0.005 mm, Δy = 0.003 mm, and Δz = 0.002 mm, with a comprehensive offset deviation ΔP ≈ 0.006 mm. Based on a preset correction threshold of ±0.01mm, deviations exceeding or approaching the threshold are corrected. The coordinate values ​​of the three-dimensional shape data are adjusted through a coordinate transformation algorithm. After correction, the overall offset deviation of defect 1 is reduced to 0.003mm, while defects 2 and 3 maintain their original low deviation levels. The calibrated three-dimensional shape data is obtained, and the data accuracy meets the high-precision requirements of industrial inspection.

[0046] Subsequently, the matching and verification of signal features and three-dimensional shape features are performed: The calibrated three-dimensional shape data is extracted, including the core features such as the size parameters, contour curves, and spatial positions of each defect. Simultaneously, leakage magnetic signal segments corresponding to the defect regions are selected from the initial signal set in step S11 (defect 1 corresponds to a signal segment with an axial direction of 2.3-2.5m and a circumferential angle of 90-100°; defect 2 corresponds to a signal segment with an axial direction of 5.1-5.3m and a circumferential angle of 220-230°; defect 3 corresponds to a signal segment with an axial direction of 7.8-8.0m and a circumferential angle of 310-320°). Thirty parameters, such as the peak value of the leakage magnetic signal amplitude and the rate of change of the magnetic field gradient, are selected for the signal features. Thirty parameters, such as the defect depth and contour curvature, are selected for the three-dimensional shape features. Correlation is calculated one-to-one, and a correlation analysis algorithm is used to calculate the matching degree between the signal features and the three-dimensional shape features. The matching degree calculation formula is as follows: Where R is the matching degree (ranging from -1 to 1, with values ​​closer to 1 indicating a higher matching degree), and S... i Let i be the characteristic parameter of the i-th signal. F represents the mean of the signal characteristic parameters. i Let i be the i-th 3D shape feature parameter. is the mean value of three-dimensional shape feature parameters, and n is the number of feature parameters (n = 30 in this embodiment). The preset matching degree threshold of 0.9 is set based on the signal-shape matching experiments of 100 groups of standard defect samples. When the matching degree ≥ 0.9, the fitting degree between the shape restoration and the actual defect reaches more than 95%, meeting the reliability requirements of industrial inspection. After calculation, the matching degree R of defect 1 is 0.98, the matching degree R of defect 2 is 0.97, and the matching degree R of defect 3 is 0.96, all higher than the preset matching degree threshold of 0.9. It is determined that the verification is qualified, indicating that the three-dimensional shape restoration highly fits the characteristics of the original magnetic flux leakage signal and there is no morphological distortion that deviates from the actual situation.

[0047] Finally, image generation and output are performed: Based on the three-dimensional shape data that passed the verification, start the three-dimensional imaging rendering engine and set the core parameters for image generation: imaging resolution 512×512×256 pixels, point cloud density 0.01mm / point, color mapping rule (assign colors according to the defect depth, the deeper the depth, the closer the color is to red, and the shallower the depth, the closer the color is to blue), and lighting rendering mode (global illumination + local specular highlight). Use the improved Poisson surface reconstruction algorithm, which can quickly generate a smooth and continuous surface model based on discrete grid data, adapt to the restoration of the minute details of defects, and ensure the clarity and fidelity of image generation. Convert the three-dimensional shape data into a visual grid model through the surface reconstruction algorithm. After texture mapping and color rendering, generate a high-precision three-dimensional imaging result that supports rotation, scaling, and剖切 viewing. The output format is a dual format of STL (for industrial modeling analysis) and PLY (for visual display). The generated three-dimensional imaging can clearly present the accurate positions (axial, circumferential, and radial coordinates) of the 3 defects, the complete morphology (contour details of elliptical corrosion pits, linear cracks, and pitting corrosion), and the key dimensions (major axis, minor axis, length, width, diameter, depth, etc.). Verified by a third-party laser three-dimensional scanner, the imaging size measurement error ≤ 0.03mm, and the position deviation ≤ 0.02mm, fully meeting the core requirements of high-precision imaging for industrial pipeline defect detection and providing an intuitive and reliable digital basis for subsequent defect risk assessment and repair plan formulation.

[0048] To sum up, the high-precision three-dimensional imaging method based on continuous magnetic flux leakage data disclosed in this invention is oriented towards the core requirements in the fields of industrial inspection and non-destructive evaluation, and constructs a complete technical closed-loop. The method accurately extracts the feature vectors of the magnetic flux leakage signal through a convolutional neural network model, breaks through the positioning ambiguity bottleneck by combining the iterative mapping algorithm, optimizes the morphological data through geometric boundary mining and shape error correction, and finally generates a high-fidelity three-dimensional imaging of defects through grid reconstruction, hole filling, and two-dimensional verification (precision calibration + signal matching).

[0049] The embodiments, using a DN500mm industrial carbon steel pipe as the inspection object, verified the feasibility and reliability of the method: imaging position deviation ≤0.02mm, dimensional measurement error ≤0.03mm, and signal-to-morphological feature matching degree ≥0.96, significantly better than traditional magnetic flux leakage imaging technology. This invention effectively solves the problems of fuzzy positioning, morphological distortion, and insufficient adaptability of traditional methods, ensuring both the accuracy and integrity of image generation and enhancing the adaptability to complex industrial scenarios. It provides reliable technical support for defect detection of critical infrastructure such as pipelines and bridges, and promotes the development of industrial non-destructive testing towards high precision, digitalization, and visualization.

[0050] refer to Figure 2 The second embodiment of the invention provides a high-precision three-dimensional imaging system based on continuous magnetic flux leakage data, comprising: Signal feature extraction module: Acquires the surface magnetic flux leakage signal of the target under test through a magnetic flux leakage signal acquisition device to form an initial signal set, and extracts the feature vector of the initial signal set using a preset convolutional neural network model; Three-dimensional preliminary positioning module: Separates three-dimensional spatial data from the initial signal set corresponding to the feature vector, and calculates the preliminary three-dimensional position coordinates of the corresponding defect by using a preset iterative mapping algorithm for the positioning ambiguity area in the three-dimensional spatial data; Refined position modeling module: Based on the preliminary three-dimensional position coordinates, the spatial geometric features of the defect are mined and the potential geometric boundaries of the defect are determined. When the deviation between the potential geometric boundary and the surface leakage magnetic signal is less than a preset deviation threshold, the boundary data corresponding to the potential geometric boundary is fused to construct a refined position model. Shape error correction module: Based on the refined position model, extract the detection subset of defects. When the shape error in the detection subset exceeds the preset allowable range, apply the preset correction function to correct the detection subset and obtain the corrected shape parameters. Mesh Reconstruction and Filling Module: Based on the corrected shape parameters, a spatial reconstruction mesh of the defect is constructed, and missing information is filled into the void areas in the spatial reconstruction mesh to obtain a complete three-dimensional shape representation of the defect; 3D calibration imaging module: Based on the complete 3D shape representation, it performs 3D accuracy calibration and leakage magnetic signal matching verification. After the verification is qualified, it outputs a high-precision 3D image of the defect. The detection subset includes local spatial coordinate data, local geometric contour data, local magnetic flux leakage signal characteristic data, and local morphological deviation data of the defect.

[0051] It should be noted that the high-precision three-dimensional imaging system based on continuous magnetic flux leakage data provided in this embodiment of the invention is used to execute all the process steps of the high-precision three-dimensional imaging method based on continuous magnetic flux leakage data in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0052] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A high-precision three-dimensional imaging method based on continuous magnetic flux leakage data, characterized in that, include: The surface magnetic flux leakage signal of the target under test is acquired by a magnetic flux leakage signal acquisition device to form an initial signal set, and the feature vector of the initial signal set is extracted by a preset convolutional neural network model. Three-dimensional spatial data is separated from the initial signal set corresponding to the feature vector, and the preliminary three-dimensional position coordinates of the corresponding defects are calculated by a preset iterative mapping algorithm for the positioning ambiguity areas in the three-dimensional spatial data. Based on the preliminary three-dimensional position coordinates, the spatial geometric features of the defect are mined and the potential geometric boundary of the defect is determined. When the deviation between the potential geometric boundary and the surface leakage magnetic signal is less than a preset deviation threshold, the boundary data corresponding to the potential geometric boundary is fused to construct a refined position model. Based on the refined position model, a detection subset of defects is extracted. When the shape error in the detection subset exceeds a preset allowable range, a preset correction function is applied to correct the detection subset to obtain the corrected shape parameters. Based on the corrected shape parameters, a spatial reconstruction mesh of the defect is constructed, and missing information is filled into the empty areas in the spatial reconstruction mesh to obtain a complete three-dimensional shape representation of the defect. Based on the complete three-dimensional shape representation, three-dimensional accuracy calibration and leakage magnetic signal matching verification are performed. After the verification is qualified, a high-precision three-dimensional image of the defect is output. The detection subset includes local spatial coordinate data, local geometric contour data, local magnetic flux leakage signal characteristic data, and local morphological deviation data of the defect.

2. The high-precision three-dimensional imaging method based on continuous magnetic flux leakage data according to claim 1, characterized in that, The process of acquiring the surface magnetic flux leakage signal of the target under test through a magnetic flux leakage signal acquisition device to form an initial signal set, and extracting the feature vector of the initial signal set using a preset convolutional neural network model, includes: The magnetic flux leakage signal acquisition device is attached to the surface of the target under test to a preset accuracy, and the magnetic flux leakage signal of the entire surface of the target under test is acquired at a preset frequency to obtain the original magnetic flux leakage signal. The original magnetic flux leakage signal is processed by removing invalid signals caused by interference and noise, and then organized according to the acquisition time sequence to form an initial signal set. The initial signal set is input into a preset convolutional neural network model, and the leakage magnetic signal features are extracted and compressed layer by layer through convolutional layers and pooling layers; The leakage magnetic signal features are fused through the fully connected layer of the convolutional neural network model to output a feature vector representing the correlation between the leakage magnetic signal and the defect.

3. The high-precision three-dimensional imaging method based on continuous magnetic flux leakage data according to claim 1, characterized in that, The step of separating three-dimensional spatial data from the initial signal set corresponding to the feature vector, and calculating the preliminary three-dimensional position coordinates of the corresponding defects using a preset iterative mapping algorithm for the positioning ambiguity regions existing in the three-dimensional spatial data, includes: By associating the feature vectors with the initial signal set, three-dimensional spatial data related to defect localization are extracted from the initial signal set; The three-dimensional spatial data is scanned across the entire domain to identify corresponding ambiguous and normal regions, and abnormal data points and data distribution characteristics within the ambiguous regions are marked. The marked ambiguous positioning area and the normal area are input into a preset iterative mapping algorithm, and calibration is performed through iterative mapping to obtain calibrated three-dimensional spatial data; Based on the calibrated three-dimensional spatial data, the preliminary three-dimensional location coordinates of the defect are calculated.

4. The high-precision three-dimensional imaging method based on continuous magnetic flux leakage data according to claim 1, characterized in that, Based on the preliminary three-dimensional position coordinates, the spatial geometric features of the defect are mined, and the potential geometric boundaries of the defect are determined. When the deviation between the potential geometric boundaries and the surface magnetic flux leakage signal is less than a preset deviation threshold, the boundary data corresponding to the potential geometric boundaries are fused to construct a refined position model, including: Based on the preliminary three-dimensional position coordinates, the leakage magnetic signal features of the corresponding area are extracted, and the spatial geometric features of the defects are mined based on the leakage magnetic signal features. Based on the spatial geometric features, the potential geometric boundaries of the defects are determined, and the boundary data corresponding to the potential geometric boundaries are extracted. Calculate the signal deviation between the potential geometric boundary and the surface leakage magnetic field signal, and compare the signal deviation with a preset deviation threshold; When the deviation is less than a preset deviation threshold, the boundary data is fused to construct a refined position model.

5. A high-precision three-dimensional imaging method based on continuous magnetic flux leakage data according to claim 1, characterized in that, Based on the refined position model, a detection subset of defects is extracted. When the shape error in the detection subset exceeds a preset allowable range, a preset correction function is applied to correct the detection subset to obtain corrected shape parameters, including: Extract a subset of detectors related to defect morphology from the refined location model; The probe subset is subjected to morphological analysis to detect existing shape errors and calculate shape error values; wherein, the shape errors include local distortion deviation of defect contour, local error of defect size, local offset deviation of defect boundary and local distortion deviation of defect morphology. The shape error value is compared with a preset allowable range; When the shape error value exceeds the preset allowable range, a preset correction function is invoked to correct the shape error and obtain the corrected shape parameters.

6. The high-precision three-dimensional imaging method based on continuous magnetic flux leakage data according to claim 1, characterized in that, The process of constructing a spatial reconstruction mesh of the defect based on the corrected shape parameters, and filling missing information in the void regions of the spatial reconstruction mesh to obtain a complete three-dimensional shape representation of the defect includes: Based on the corrected shape parameters, set the mesh construction parameters and construct a spatial reconstruction mesh for the defects; Perform global detection on the spatial reconstruction mesh to identify and mark the range of void regions; Extract the magnetic flux leakage signal and geometric feature data corresponding to the void region, and fill in the missing information of the void region; After filling in the missing information, a complete three-dimensional shape representation of the defect is obtained.

7. The high-precision three-dimensional imaging method based on continuous magnetic flux leakage data according to claim 1, characterized in that, Based on the complete three-dimensional shape representation, three-dimensional accuracy calibration and magnetic flux leakage signal matching verification are performed. After the verification is passed, a high-precision three-dimensional image of the defect is output, including: The complete three-dimensional shape representation of the defect is invoked, and a three-dimensional accuracy calibration operation is performed based on the preset calibration standard. During the calibration process, the spatial offset deviation is corrected simultaneously to obtain the calibrated three-dimensional shape data. Extract the calibrated three-dimensional shape data, select the leakage magnetic signal segment corresponding to the initial signal set, perform matching verification between signal features and three-dimensional shape features, and output a high-precision three-dimensional image of the defect.

8. A high-precision three-dimensional imaging system based on continuous magnetic flux leakage data, characterized in that, include: Signal feature extraction module: Acquires the surface magnetic flux leakage signal of the target under test through a magnetic flux leakage signal acquisition device to form an initial signal set, and extracts the feature vector of the initial signal set using a preset convolutional neural network model; Three-dimensional preliminary positioning module: Separates three-dimensional spatial data from the initial signal set corresponding to the feature vector, and calculates the preliminary three-dimensional position coordinates of the corresponding defect by using a preset iterative mapping algorithm for the positioning ambiguity area in the three-dimensional spatial data; Refined position modeling module: Based on the preliminary three-dimensional position coordinates, the spatial geometric features of the defect are mined and the potential geometric boundaries of the defect are determined. When the deviation between the potential geometric boundary and the surface leakage magnetic signal is less than a preset deviation threshold, the boundary data corresponding to the potential geometric boundary is fused to construct a refined position model. Shape error correction module: Based on the refined position model, extract the detection subset of defects. When the shape error in the detection subset exceeds the preset allowable range, apply the preset correction function to correct the detection subset and obtain the corrected shape parameters. Mesh Reconstruction and Filling Module: Based on the corrected shape parameters, a spatial reconstruction mesh of the defect is constructed, and missing information is filled into the void areas in the spatial reconstruction mesh to obtain a complete three-dimensional shape representation of the defect; 3D calibration imaging module: Based on the complete 3D shape representation, it performs 3D accuracy calibration and leakage magnetic signal matching verification. After the verification is qualified, it outputs a high-precision 3D image of the defect. The detection subset includes local spatial coordinate data, local geometric contour data, local magnetic flux leakage signal characteristic data, and local morphological deviation data of the defect.

9. A high-precision three-dimensional imaging system based on continuous magnetic flux leakage data according to claim 8, characterized in that, The three-dimensional preliminary positioning module includes: Data association and separation unit: associates the correspondence between the feature vector and the initial signal set, and separates the three-dimensional spatial data related to defect location from the initial signal set; Full-domain scanning and marking unit: performs full-domain scanning on the three-dimensional spatial data, identifies corresponding ambiguous and normal regions, and marks abnormal data points and data distribution characteristics within the ambiguous regions; Iterative mapping calibration unit: The marked positioning fuzzy area and the normal area are input into a preset iterative mapping algorithm, and calibration is performed through iterative mapping to obtain calibrated three-dimensional spatial data; Three-dimensional coordinate calculation unit: Based on the calibrated three-dimensional spatial data, the preliminary three-dimensional position coordinates of the defect are calculated.

10. A high-precision three-dimensional imaging system based on continuous magnetic flux leakage data according to claim 9, characterized in that, The shape error correction module includes: Detector subset extraction unit: Extracts a detector subset related to the defect morphology from the refined location model; Shape error detection unit: performs shape analysis on the detection subset, detects existing shape errors and calculates shape error values; wherein, the shape errors include local distortion deviation of defect contour, local error of defect size, local offset deviation of defect boundary and local distortion deviation of defect shape; Error range comparison unit: compares the shape error value with a preset allowable range; Shape parameter correction unit: When the shape error value exceeds the preset allowable range, a preset correction function is called to correct the shape error and obtain the corrected shape parameters.

Images