High-voltage cable surface defect detection method and system based on adaptive rolling ball method

Abstract

The present application relates to a kind of high-voltage cable surface defect detection method and system based on adaptive ball method, comprising: obtaining the point cloud data of the high-voltage cable surface to be measured;Adopt ball method to process point cloud data, the ball method in the process of processing each three-dimensional point cloud region, by calculating the local point density and vertex average curvature of current three-dimensional point cloud region, determine dynamic ball radius to carry out point cloud reconstruction;To the three-dimensional mesh model obtained by point cloud reconstruction, by extracting the circumferential curvature sequence along the cable axial direction, and with the help of pre-trained gap prediction model, the gap position is predicted, and the predicted gap position is completed, to improve the morphological restoration accuracy of gap position, obtain the final high-voltage cable surface defect detection result.Compared with prior art, the present application improves the defect detection sensitivity, reduces the defect detection error.

Description

Technical Field

[0001] This invention relates to the field of cable surface defect detection technology, and in particular to a method and system for detecting surface defects in high-voltage cables based on the adaptive rolling ball method. Background Technology

[0002] In the manufacturing process of high-voltage cable joints, the surface processing accuracy of the insulation shield directly affects the uniformity of the electric field distribution and long-term operational reliability of the joint. Traditional quality inspection mainly relies on manual measurement using micrometers for discrete points, which is not only inefficient (taking more than 30 minutes per joint) but also unable to capture continuous circumferential defects, resulting in a high risk of missed defects (up to 15%). With the application of 3D laser scanning technology, automated inspection methods based on point cloud data have gradually become mainstream. However, limited by the accuracy bottleneck of surface reconstruction algorithms, existing systems are unable to meet the ±0.15mm diameter tolerance inspection requirements of high-voltage cable joint processes.

[0003] Current mainstream Poisson reconstruction algorithms face significant technical obstacles when processing cable scanning point clouds. This algorithm generates closed surfaces through global implicit function fitting, but its smoothing properties can obscure subtle local defects. Experiments show that it cannot accurately characterize diameter variations exceeding 0.38 mm within a 5 mm × 5 mm grid. Particularly in the transition section of cable joints, where the point cloud density drops sharply from 300 points / cm² to 50 points / cm² along the axial direction, Poisson reconstruction generates non-physical holes in sparse areas, requiring manual intervention for mesh repair, severely limiting the automation of the inspection process. Even more seriously, when V-shaped cutting defects exist on the surface, this algorithm tends to generate bridging triangular facets, leading to boundary misalignment errors exceeding 0.35 mm in subsequent unfolded images, directly affecting defect location accuracy.

[0004] While traditional ball-pivoting preserves geometric details, it exposes inherent defects in practical applications. Choosing a fixed radius presents a dilemma: a 5mm radius, while covering sparse areas, can over-smooth out 3.2mm-level microcracks on smooth surfaces; conversely, a 1mm radius, while retaining detail, leads to mesh breakage in areas with point cloud density below 80 points / cm², forcing an axial measurement interval of 5mm, which fails to effectively capture diameter gradient changes in transition sections. Actual testing shows that in transition regions with curvature changes exceeding 0.15mm⁻¹, the defect detection sensitivity of traditional ball-pivoting drops by up to 62%, severely limiting the accuracy of process quality assessment.

[0005] Existing cylindrical unfolding algorithms face multiple technical bottlenecks in engineering practice. When using the isometric mapping method for 3D to 2D unfolding, if the centerline fitting error exceeds 0.2mm, the unfolded image will exhibit periodic wavy distortion, obscuring the true defect morphology. For scanned missing areas, existing linear interpolation methods disrupt curvature continuity, resulting in artificial intervention marks on the completed diameter curve, with a measured notch completion accuracy of only 82.5%. More critically, traditional detection systems rely on a visual interpretation mode with 5mm×5mm color blocks, which cannot identify latent surface cracks at the 3.2mm×2.8mm level and generates a false alarm rate of 17.3% due to point cloud registration errors exceeding 0.25mm, making it difficult to meet the comprehensive quality control requirements of cable joints in smart grid construction.

[0006] In summary, traditional high-voltage cable surface inspection technologies have three major drawbacks:

[0007] First, the point cloud reconstruction stage suffers from accuracy loss due to insufficient algorithm adaptability. The global smoothness characteristic of Poisson reconstruction blurs local defect features, and the fixed radius rolling sphere law faces the dual contradiction of detail loss and mesh breakage in sparse and dense regions.

[0008] Secondly, there are defects in the geometric distortion and gap handling during the cylinder unfolding process. The existing isoangular mapping method causes distortion of the unfolded diagram due to the offset of the central axis, and linear interpolation completion destroys the curvature continuity.

[0009] Third, the detection sensitivity and reliability are insufficient. The 5mm×5mm grid analysis is difficult to capture sub-millimeter-level hidden cracks, and the point cloud registration error leads to a high false alarm rate of 17.3%. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the prior art, which suffers from accuracy loss due to insufficient algorithm adaptability in the point cloud reconstruction stage, and to provide a high-voltage cable surface defect detection method and system based on the adaptive rolling ball method.

[0011] The objective of this invention can be achieved through the following technical solutions:

[0012] A method for detecting surface defects in high-voltage cables based on the adaptive rolling ball method includes the following steps:

[0013] Acquire point cloud data of the surface of the high-voltage cable under test;

[0014] The rolling sphere method is used to process point cloud data. In the process of processing each three-dimensional point cloud region, the rolling sphere method determines the dynamic rolling sphere radius by calculating the local point density and average vertex curvature of the current three-dimensional point cloud region in order to reconstruct the point cloud.

[0015] The 3D mesh model obtained by point cloud reconstruction is used to extract the circumferential curvature sequence along the cable axis and predict the gap position with the help of a pre-trained gap prediction model. The predicted gap position is then filled in to improve the morphological restoration accuracy of the gap position, and the final high-voltage cable surface defect detection result is obtained.

[0016] Furthermore, the expression for calculating the dynamic rolling ball radius is:

[0017] ;

[0018] In the formula, For the dynamic rolling ball radius, As a density adjustment factor, As the reference radius, This is the curvature adjustment coefficient. The mean curvature of the vertices. The global average curvature, The density adjustment factor is the standard deviation of curvature. Based on local point density Values ​​are dynamically allocated.

[0019] Furthermore, the density adjustment factor Based on local point density The dynamic allocation of values ​​is specifically as follows:

[0020] when When the number of points per cm² is 1, β =0.8;

[0021] When 50≤ρ≤200 points / cm² β =1.0;

[0022] When ρ < 50 points / cm² β =1.5.

[0023] Furthermore, during the rolling ball propulsion process, a topological relationship table of edges and points is constructed to constrain the number of adjacent faces of each edge to no more than 2.

[0024] Furthermore, the gap prediction model receives an axially continuous multi-layer circumferential curvature Fourier coefficient matrix, and after extracting spatiotemporal features through a bidirectional LSTM unit, generates a gap probability heatmap.

[0025] The portion of the gap probability that is greater than a preset probability threshold is used as the predicted gap location.

[0026] Furthermore, the specific steps for filling in the predicted gap positions are as follows:

[0027] Three complete circumferential lines are selected on each side of the gap location as boundary conditions. By minimizing the energy function of the corresponding region, the completion curve is solved to complete the predicted gap location.

[0028] Furthermore, the method also includes parametrically unfolding the three-dimensional mesh model at the gap filling location, the parametric unfolding process including:

[0029] The cylinder's central axis is iteratively fitted based on a 3D mesh model, and outliers are removed.

[0030] A reference circumference is generated at preset intervals along the central axis of the cylinder. The circumference of the reference circumference is divided into multiple sampling points, and the moving least squares method is used to correct the local distortion caused by scanning noise.

[0031] By transforming the three-dimensional points into a two-dimensional parameter space, a two-dimensional unfolded plane is obtained.

[0032] Furthermore, during the polar coordinate transformation, multiple intermediate circular lines are inserted into the transition segment, and adaptive resampling is performed to increase the sampling density of the two-dimensional unfolded plane in the gradient region of the transition segment.

[0033] Furthermore, the method employs a CUDA-accelerated parallel computing module to solve for the Gaussian curvature of the vertices using the 15-neighborhood Jet-Fitting method.

[0034] The present invention also provides a high-voltage cable surface defect detection system based on the adaptive rolling ball method, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of the method described above.

[0035] Compared with the prior art, the present invention has the following advantages:

[0036] (1) This invention addresses the problem of detail loss in the transition region caused by global smoothing in the traditional Poisson reconstruction algorithm. It proposes a dynamic rolling ball radius adjustment mechanism with local curvature-density coupling, which calculates the point density ρ and curvature in real time within a 3mm neighborhood. By dynamically adjusting the rolling ball radius R, the mesh coverage in sparse areas (ρ<50 points / cm²) is increased to 98%, while the redundancy of triangular patches in dense areas (ρ>200 points / cm²) is reduced by 35%. Experimental data show that this strategy reduces the diameter fitting error from 0.38mm in the traditional method to 0.22mm. Especially in the transition section where the curvature change rate exceeds 0.15mm⁻¹, the defect detection sensitivity breaks through to the 100μm level, which is 3 times higher than the fixed radius rolling ball method.

[0037] (2) For the prediction of gaps on the surface of high-voltage cables, the present invention adopts a bidirectional LSTM gap prediction model to achieve high recall recognition by analyzing the Fourier descriptor features of five consecutive circumferential curvature layers in the axial direction; and completes the predicted gap position to improve the morphological restoration accuracy of the gap position; during the completion process, combined with the cubic B-spline energy minimization interpolation algorithm, the morphological restoration accuracy of the gap completion can reach 95.7%, which is significantly improved compared with the traditional linear interpolation method, and makes the boundary alignment error of the completion area strictly controlled within a very small range, eliminating the need for manual intervention while greatly improving the efficiency of the detection process.

[0038] (3) The present invention uses a parallel computing module based on CUDA acceleration to solve the Gaussian curvature of the vertex by the 15-neighbor Jet-Fitting method, which can reduce the curvature field calculation time to 1 / 8 of the traditional CPU solution; combined with the local density sensing capability of the dynamic radius adjustment strategy, the complete detection cycle of a single 1.5-meter cable is shortened to less than 3 minutes, the amount of grid data is reduced by 35%, and the memory usage is reduced to 58% of the traditional rolling ball method.

[0039] (4) For the parametric unfolding of the three-dimensional mesh model, the present invention adopts an improved cylindrical parametric unfolding algorithm, which completely eliminates the wavy distortion of the traditional isoangular mapping method by using RANSAC centerline fitting (residual threshold 0.1mm) and adaptive resampling technology, thereby reducing the geometric distortion rate of the unfolded diagram by 65.7%. Attached Figure Description

[0040] Figure 1 This is a schematic flowchart of a high-voltage cable surface defect detection method based on the adaptive rolling ball method provided in an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram illustrating the principle of adaptive rolling ball radius dynamic adjustment provided in an embodiment of the present invention;

[0042] Figure 3 This is a flowchart of a gap prediction and completion algorithm provided in an embodiment of the present invention;

[0043] Figure 4 This is a geometric mapping diagram of a cylindrical parametric unfolding algorithm provided in an embodiment of the present invention;

[0044] Figure 5 This is a comparison chart of defect visualization effects provided in an embodiment of the present invention;

[0045] Figure 6 This is an architecture diagram of a high-voltage cable surface defect detection method based on the adaptive rolling ball method provided in an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0047] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0048] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0049] Example 1

[0050] like Figure 1 As shown, this embodiment provides a method for detecting surface defects in high-voltage cables based on the adaptive rolling ball method, including the following steps:

[0051] S1: Acquire point cloud data of the surface of the high-voltage cable under test;

[0052] S2: The rolling sphere method is used to process the point cloud data. In the process of processing each three-dimensional point cloud region, the rolling sphere method calculates the local point density and the average curvature of the vertices of the current three-dimensional point cloud region to determine the dynamic rolling sphere radius for point cloud reconstruction.

[0053] S3: For the 3D mesh model obtained by point cloud reconstruction, the circumferential curvature sequence is extracted along the cable axis, and the gap position is predicted with the help of a pre-trained gap prediction model. The predicted gap position is then filled in to improve the morphological restoration accuracy of the gap position, and the final high-voltage cable surface defect detection result is obtained.

[0054] Specifically, in step S1, this embodiment uses a 26-line cross-blue laser scanner to acquire point cloud data of the cable surface.

[0055] In step S2, this embodiment optimizes the rolling sphere reconstruction process through a dynamic coupling mechanism between local curvature and point cloud density. For each 3D point cloud region to be processed, the KD-Tree index is used to quickly retrieve the point set within a 3mm radius neighborhood, and the local point density ρ (unit points / cm²) and the vertex average curvature calculated based on the Jet-Fitting method are calculated simultaneously. .

[0056] like Figure 2 As shown, the expression for calculating the dynamic rolling ball radius is:

[0057] ;

[0058] In the formula, For the dynamic rolling ball radius, As a density adjustment factor, As the reference radius, This is the curvature adjustment coefficient. The mean curvature of the vertices. The global average curvature, Standard deviation of curvature, density adjustment factor Based on local point density Values ​​are dynamically allocated.

[0059] In this embodiment, the reference radius Set to 2mm, curvature adjustment coefficient α=0.6, density adjustment factor β Dynamic allocation is based on the ρ value: when ρ>200, β=0.8 to suppress excessive smoothing and preserve detailed features; when 50≤ρ≤200, β=1.0 to maintain standard processing; when ρ<50, β=1.5 to expand the coverage area and ensure topological continuity. During the rolling ball propagation process, an edge-point topology table is constructed simultaneously. By constraining the number of adjacent faces for each edge to not exceed 2, mesh self-intersection is effectively avoided.

[0060] In step S3, the gap prediction model receives the axial continuous multi-layer circumferential curvature Fourier coefficient matrix, and after extracting the spatiotemporal features through a bidirectional LSTM unit, it generates a gap probability heatmap.

[0061] The portion of the gap probability that is greater than a preset probability threshold is used as the predicted gap location.

[0062] Specifically, such as Figure 3 As shown, in this embodiment, after completing the initial three-dimensional mesh reconstruction, the system extracts the circumferential curvature sequence along the cable axis at 10mm intervals and predicts the gap location based on a bidirectional LSTM network. The network input layer receives the first 8 coefficients of the Fourier descriptors from 5 consecutive layers (axial positions t-2 to t+2). After the spatiotemporal features are extracted by a 64-element bidirectional LSTM layer, the gap occurrence probability P_gap is calculated by the Sigmoid output layer, generating a gap probability heatmap.

[0063] That is, the gap prediction model includes:

[0064] Input layer: The first 8 coefficients of the Fourier descriptor of the circumferential curvature of the input axis are 5 consecutive layers.

[0065] LSTM layer: Bidirectional 64-unit long short-term memory network, with a time step size of 5;

[0066] Output layer: The Softmax classifier predicts the probability of gap occurrence, and the recall rate reaches 92.4% when the threshold is set to 0.78.

[0067] The specific steps for filling in the predicted gaps are as follows:

[0068] Three complete circumferential lines are selected on each side of the gap location as boundary conditions. By minimizing the energy function of the corresponding region, the completion curve is solved to complete the predicted gap location.

[0069] In this embodiment, when P_gap exceeds a preset threshold of 0.78, a cubic uniform B-spline surface completion algorithm is triggered: three complete circular lines are selected on each side of the gap axis as boundary conditions, and the energy function is minimized. :

[0070] ;

[0071] In the formula, The complete curve to be solved. For the boundary data points, there are three complete circumferences selected on each side of the gap axis. To complete the curve The second derivative, To complete the curve The parameter, given a The value corresponds to the completed curve. The coordinates of a point on the graph;

[0072] Solving for the complete curve This ensures that the completed surface meets C1 continuity, improving the shape restoration accuracy of the gap region to 95.7%.

[0073] Preferably, the method further includes:

[0074] S4: Parametrically unfold the 3D mesh model at the missing area. This parametric unfolding process includes:

[0075] The cylinder's central axis is iteratively fitted based on a 3D mesh model, and outliers are removed.

[0076] A reference circumference is generated at preset intervals along the central axis of the cylinder. The circumference of the reference circumference is divided into multiple sampling points, and the moving least squares method is used to correct the local distortion caused by scanning noise.

[0077] By transforming the three-dimensional points into a two-dimensional parameter space, a two-dimensional unfolded plane is obtained.

[0078] In this embodiment, as Figure 4 and 5 As shown, the cylinder's central axis is first iteratively fitted using the RANSAC algorithm, and outliers are removed by setting a residual threshold of 0.1 mm. Then, a baseline circumference is generated along the axis at 10 mm intervals, dividing the circumference into 1024 equal sampling points. Moving least squares is then used to correct local distortions caused by scanning noise. During the polar coordinate transformation stage, the three-dimensional point (x, y, z) is mapped to the two-dimensional parameter space (θ, l), where θ = arctan(y / x) achieves circumferential expansion, and l is the cumulative axial length.

[0079] Preferably, during the polar coordinate transformation process, multiple intermediate circular lines are inserted into the transition segment, and adaptive resampling is performed to increase the sampling density of the two-dimensional unfolded plane in the gradient region of the transition segment.

[0080] In this embodiment, for the transition section curvature abrupt change region (ΔD / Δl>0.15mm / mm), 5 layers of intermediate circumferential lines are inserted, and the sampling density of the unfolded image in the gradient region is increased by 3 times through adaptive resampling, effectively suppressing geometric distortion.

[0081] Preferably, the method employs a CUDA-accelerated parallel computing module to solve the Gaussian curvature of the vertices using the 15-neighborhood Jet-Fitting method, reducing the computation time to 1 / 8 of that of traditional methods.

[0082] For the display process of the two-dimensional unfolded plane, this embodiment maps the diameter deviation ΔD to the HSV color space, setting the H channel to correspond to 240° (blue) when ΔD=+0.5mm and 0° (red) when ΔD=-0.5mm. A 2.5mm×2.5mm color block matrix is ​​generated by combining Gaussian kernel density estimation (bandwidth h=3mm). The improved ICP registration algorithm introduces a curvature similarity constraint term, controlling the alignment error between the 3D Mesh model and the unfolded diagram to within 0.12mm. The final generated PDF inspection report integrates the 3D model, axial heat map, and circumferential defect distribution curve, supporting digital backtracking and optimization of process parameters.

[0083] In terms of visualization, the improved cylindrical parametric unfolding algorithm completely eliminates the wavy distortion of the traditional isogonal mapping method through RANSAC central axis fitting (residual threshold of 0.1mm) and adaptive resampling technology, reducing the geometric distortion rate of the unfolded image by 65.7%. Combined with a 2.5mm×2.5mm color block matrix and HSV chromatographic mapping (H channel 240°~0° corresponds to ±0.5mm diameter deviation), the system can clearly present the diameter gradient change at 10mm intervals in the axial direction and the defect distribution with a 5° resolution in the circumferential direction. The measured X-marker defect location accuracy reached 98.2%, and the false alarm rate dropped sharply from 17.3% to 4.1%. It also successfully identified a 3.2mm×2.8mm hidden crack, breaking through the industry's detection limit of 5mm×5mm.

[0084] Regarding data security and traceability, this solution employs a fully localized processing workflow to avoid the external transmission of point cloud data. Combined with AES-256 encrypted PDF report storage and 0.12mm-level spatial registration accuracy between the 3D mesh and the unfolded diagram, it constructs an immutable digital twin archive. Compared to existing solutions that rely on cloud-based analysis, this solution reduces data transmission volume by 75% and compresses the original point cloud storage requirement to 1.2GB / km (compared to 2.8GB for traditional solutions) through an adaptive sampling mechanism, significantly reducing storage and transmission costs. This invention, through the deep integration of algorithmic innovation and hardware acceleration, achieves synergistic optimization of detection accuracy, processing efficiency, resource utilization, and security, establishing a fully digital and highly robust quality assessment system for high-voltage cable joint processes.

[0085] like Figure 6 As shown, starting from the laser scanner data input, the system sequentially connects to the dynamic radius reconstruction module (marked with the CUDA acceleration icon), the notch processing engine, the parametric unfolding unit, and the visualization platform. The data flow arrows indicate the interaction protocols between the modules. A partial view of the PDF report sample is inserted in the lower right corner, showing the linkage annotation effect between the axial heat map and the 3D model.

[0086] This solution, through the seamless integration of the aforementioned technical links, maintains a reconstruction accuracy of 0.25mm while increasing the defect detection sensitivity of the 5mm×5mm grid to the 100μm level and reducing the false alarm rate to 4.1%, thus achieving a comprehensive quantitative assessment of the surface quality of high-voltage cable joints.

[0087] Example 2

[0088] This embodiment provides a high-voltage cable surface defect detection system based on the adaptive rolling ball method, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of the high-voltage cable surface defect detection method based on the adaptive rolling ball method as described in Embodiment 1.

[0089] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0090] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0093] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0094] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for detecting surface defects in high-voltage cables based on the adaptive rolling ball method, characterized in that, Includes the following steps: Acquire point cloud data of the surface of the high-voltage cable under test; The rolling sphere method is used to process point cloud data. In the process of processing each three-dimensional point cloud region, the rolling sphere method determines the dynamic rolling sphere radius by calculating the local point density and average vertex curvature of the current three-dimensional point cloud region in order to reconstruct the point cloud. The 3D mesh model obtained by point cloud reconstruction is used to extract the circumferential curvature sequence along the cable axis and predict the gap position with the help of a pre-trained gap prediction model. The predicted gap position is then filled in to improve the morphological restoration accuracy of the gap position and obtain the final high-voltage cable surface defect detection result. The formula for calculating the dynamic rolling ball radius is: ； In the formula, For the dynamic rolling ball radius, Density adjustment factor, As the reference radius, This is the curvature adjustment coefficient. The mean curvature of the vertices. The global average curvature, The density adjustment factor is the standard deviation of curvature. Based on local point density Values ​​are dynamically allocated; During the rolling ball propulsion process, a topological relationship table of edges and points is constructed to constrain the number of adjacent faces of each edge to not exceed 2. The density adjustment factor Based on local point density The dynamic allocation of values ​​is specifically as follows: when dot / cm 2 hour, β =0.8; When 50≤ρ≤200 points / cm 2 hour, β =1.0; When ρ < 50 points / cm 2 hour, β =1.5; The method further includes parametric unfolding of the 3D mesh model at the gap filling location, the parametric unfolding process including: The cylinder's central axis is iteratively fitted based on a 3D mesh model, and outliers are removed. A reference circumference is generated at preset intervals along the central axis of the cylinder. The circumference of the reference circumference is divided into multiple sampling points, and the moving least squares method is used to correct the local distortion caused by scanning noise. By transforming the three-dimensional points into the two-dimensional parameter space, a two-dimensional unfolded plane is obtained; The specific steps for filling in the predicted gap positions are as follows: Three complete circumferential lines are selected on each side of the gap location as boundary conditions. By minimizing the energy function of the corresponding region, the completion curve is solved to complete the predicted gap location.

2. The method for detecting surface defects in high-voltage cables based on the adaptive rolling ball method according to claim 1, characterized in that, The gap prediction model receives an axially continuous multi-layer circumferential curvature Fourier coefficient matrix, and after extracting spatiotemporal features through a bidirectional LSTM unit, generates a gap probability heatmap. The portion of the gap probability that is greater than a preset probability threshold is used as the predicted gap location.

3. The method for detecting surface defects in high-voltage cables based on the adaptive rolling ball method according to claim 1, characterized in that, During the polar coordinate transformation, multiple intermediate circular lines are inserted into the transition segment, and adaptive resampling is performed to increase the sampling density of the two-dimensional unfolded plane in the gradient region of the transition segment.

4. The method for detecting surface defects in high-voltage cables based on the adaptive rolling ball method according to claim 1, characterized in that, The method employs a CUDA-accelerated parallel computing module to solve for the Gaussian curvature of vertices using the 15-neighborhood Jet-Fitting method.

5. A high-voltage cable surface defect detection system based on the adaptive rolling ball method, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor calling the computer program to perform the steps of the method as described in any one of claims 1 to 4.

Images