A cable joint robust reference axis modeling and point cloud segmentation method

Abstract

The application provides a cable joint robust reference axis modeling and point cloud segmentation method, relates to the technical field of cable joint detection, obtains multi-view point cloud, equipment calibration data, cable structure templates and stripping and cutting process templates, establishes measurement credible weight and non-twisting curve coordinate system; constructs a virtual unstripped cable matrix according to the reference axis, the local cross-section deformation matrix, the nested material layer contour and the direction-related axial termination boundary, associates the point cloud points to the material layer longitudinal surface, the cut surface or the non-matrix abnormal state, jointly inverts the reference axis, the material layer contour, the stripping boundary and the point generation source label; hides and reconstructs the calibration matrix parameters through the stable axial segment, distinguishes the real stripping, the scanning shielding and the indeterminable area according to the material layer retention state, and determines the supplementary scanning view angle according to the parameter information increment, and outputs the reference axis, the layered point cloud, the cut boundary and the original point index.

Description

Technical Field

[0001] This invention relates to the field of cable joint inspection technology, specifically a method for robust reference axis modeling and point cloud segmentation of cable joints. Background Technology

[0002] The fabrication process of power cable intermediate and terminal joints includes outer sheath removal, metal shielding treatment, semi-conductive layer removal, insulation treatment, and conductor exposure. After construction, the cable joint forms a stepped structure along the axial direction, with multiple material layers terminating sequentially. Cable joint acceptance involves the exposed core length, insulation break location, insulation diameter, semi-conductive layer stripping location, outer sheath removal quality, and the consistency of the cut circumference. Traditional acceptance relies on rulers, tape measures, and calipers for localized measurements on-site. Measurement results are affected by the placement of the measuring tools, measurement angle, contact condition, and operator judgment. Maintaining a consistent measurement standard is difficult when the same joint is repeatedly measured by different personnel.

[0003] 3D scanning equipment can convert the surface of a cable joint into point cloud data containing spatial coordinates, color, reflection intensity, and normal information. However, existing point cloud processing methods typically first establish a linear principal axis based on the main direction of the entire point cloud, or perform cylindrical fitting on a complete cylindrical surface, and then form a cross-section along the obtained axis and segment the point cloud. Existing point cloud segmentation methods mainly output material categories based on color, curvature, normal vector, point density, or training models. These methods do not establish the radial nesting relationship between cable material layers or the axial step-by-step stripping relationship. When the outer layer material does not form a point cloud due to scanning obstruction, existing methods easily identify the missing point cloud as material removal; when the inner layer material forms locally visible points due to the cut slope, existing methods easily identify transition points as fully exposed material layers. Reference axis errors also change the radial distance from the point to the axis, causing the true circular cross-section to appear as an ellipse, eccentricity, or radial undulation in the incorrect coordinate system, and propagating the reference axis error to subsequent fracture location and diameter measurements. Summary of the Invention

[0004] Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a robust reference axis modeling and point cloud segmentation method for cable joints. Addressing issues such as progressive material layer stripping, varying material exposure across different axial sections, cable body bending, inconsistent cut directions, local surface defects, and scan occlusion in cable joints, this invention establishes a unified reference axis spanning all exposed material sections. Under this unified reference axis, it distinguishes between the longitudinal surface of the material layer, the surface of the material layer cut, the actual stripped area, the scan occlusion area, and non-parent body anomalies.

[0006] This invention also solves the problem of model parameters being unidentifiable due to the substitution between reference axis position, cross-sectional elliptical deformation, and material layer contour; solves the problem that point cloud local fitting results cannot be independently verified by other axial segments; and solves the problem of lacking directional supplementary sampling basis when scanning data is insufficient.

[0007] Technical solution

[0008] To achieve the above objectives, the present invention provides the following technical solution: a robust reference axis modeling and point cloud segmentation method for cable joints, comprising the following steps: acquiring the original point cloud of the cable joint, scanning equipment calibration parameters, cable structure template, and joint stripping process template; converting the original points in different scanning frames to a unified engineering coordinate system; and determining the measurement confidence weight of each point based on point density, normal consistency, scanning incident direction, multi-frame registration residual, and equipment acquisition confidence; establishing an initial reference axis based on the axial extension direction of the original point cloud; and establishing a point cloud segmentation method along the initial reference axis, consisting of axial parameters, cross-sectional orientation angle, and radial position. A torsional curve coordinate system is established; a virtual unstripped cable matrix is ​​created, comprising a spatial curve-shaped reference axis, a local cross-sectional deformation matrix distributed along the reference axis, standard cross-sectional profiles of material layers nested from the inside out, and axial termination boundaries of each material layer varying with the cross-sectional orientation angle; the retention state of each material layer at each axial position and cross-sectional orientation angle is determined based on the axial termination boundaries, and the outermost material layer in the retention state is determined as the theoretically exposed material layer at the corresponding position; a point cloud generation surface set is established based on the theoretically exposed material layers, the point cloud generation surface set including the material layers. The longitudinal surface, the cut surface at the material layer termination boundary, and the non-parent body anomaly state are analyzed. Based on the normalized distance of measurement error from each point to each generated surface in the point cloud generation surface set, material property differences, and material layer retention status, the generation source label for each point is determined. Under constraints of reference axis continuity, local cross-sectional deformation continuity, radial nesting relationship of material layers, material layer peeling order, and parameter uniqueness, the undetermined reference axis, local cross-sectional deformation matrix, standard cross-sectional profile of the material layer, axial termination boundary, and generation source label are jointly updated. Hidden verification segments are selected from stable axial segments of the same material layer, excluding the hidden verification segments. Under the condition of segment, the hidden verification segment is reconstructed based on the remaining stable axial segments, and the virtual unstripped cable parent body is updated based on the counterfactual reconstruction residual between the reconstructed position and the actual point cloud of the hidden verification segment; the original point cloud is divided into longitudinal surface point clouds of each material layer, cut surface point clouds of each material layer, and non-parent body abnormal point clouds based on the updated generation source label, and the material stripping area, scan occlusion area, and undeterminable area are distinguished based on the axial termination boundary of the outer material layer and the scanning view occlusion relationship; the reference axis verified by counterfactual reconstruction, the material layer segmented point cloud, the material layer cut boundary, and the original point index corresponding to each output data are output.

[0009] Preferably, the measurement confidence weight is determined by the product of the point density weight, the normal consistency weight, the scanning incident direction weight, the multi-frame coincidence residual weight, and the device acquisition confidence weight. The multi-frame coincidence residual is determined based on the distance from the corresponding point of the same entity surface in different scanning frames to the local fitting surface. The scanning incident direction weight is determined based on the angle between the point normal vector and the observation direction from the scanning device to that point.

[0010] Preferably, the torsion-free curvilinear coordinate system includes a reference axis unit tangent vector, a first cross-section unit basis vector, and a second cross-section unit basis vector. The first cross-section unit basis vector and the second cross-section unit basis vector propagate along the reference axis through the minimum rotation between adjacent reference axis unit tangent vectors. The parameter uniqueness constraint includes that the centroid of the standard cross-section profile of each material layer relative to the reference axis is zero, and the average area scaling of the local cross-section deformation matrix along the reference axis is zero.

[0011] Preferably, the longitudinal surface of any target material layer is formed by adding the three-dimensional coordinates of the reference axis at the axial parameter, the local section basis matrix at the corresponding position, the local section deformation matrix, and the standard section profile of the target material layer at the section direction angle; the cut surface of the target material layer is formed by the three-dimensional coordinates of the reference axis at the axial termination boundary of the target material layer, the local section basis matrix, the local section deformation matrix, and the radial interpolation between the standard section profile of the adjacent inner material layer of the target material layer and the standard section profile of the target material layer.

[0012] Preferably, the material layer retention state is determined based on the reference axis arc length parameter, the peeling advance direction, and the axial termination boundary of the material layer; at any reference axis arc length parameter and cross-sectional direction angle, the material layer that is still in the retention state and has the largest material layer number is determined as the theoretical exposed material layer, and the longitudinal surface of the material layer located inside the theoretical exposed material layer is excluded from the point cloud generation surface set at the corresponding position.

[0013] Preferably, the generation source label is determined based on the sum of the squared Mahalanobis distance from the point to the candidate generation surface, the material property difference cost, and the material layer retention state conflict cost; when the candidate generation surface does not conform to the theoretical exposed material layer relationship, the material layer retention state conflict cost is set as a prohibition cost; points that cannot be explained by the legal material layer longitudinal surface and the legal material layer cut surface are assigned to non-parent anomalous states.

[0014] Preferably, the joint update is performed cyclically in the following order: updating the source label under the conditions of fixed reference axis, local cross-sectional deformation matrix, standard cross-sectional profile of material layer, and axial termination boundary; updating the standard cross-sectional profile and local cross-sectional deformation matrix of material layer under the conditions of fixed source label and reference axis; updating the axial termination boundary of each material layer under the conditions of fixed standard cross-sectional profile and source label; establishing a set of stable points from the longitudinal surface points of material layer that are more than the protection distance from the axial termination boundary, whose orientation angle coverage reaches the coverage threshold, and whose measurement confidence weight reaches the confidence threshold, and updating the reference axis according to the set of stable points; terminating the loop when the average change of reference axis, the average change of axial termination boundary, and the change rate of source label all reach the corresponding convergence threshold.

[0015] Preferably, the hidden verification segment consists of points in the same material layer whose axial continuous length reaches a length threshold, whose orientation angle coverage reaches a coverage threshold, whose distance from the point to the longitudinal surface of the corresponding material layer reaches a residual threshold, and whose distance from the point to the axial termination boundary reaches a protection distance. After excluding the hidden verification segment, the hidden verification segment is reconstructed based on the remaining stable axial segments in the same layer, the standard cross-sectional profile of adjacent material layers, the local cross-sectional deformation matrix, and the reference axis, and the counterfactual reconstruction residual is determined based on the weighted average squared distance.

[0016] Preferably, when the parameter information matrix of the virtual unstripped cable matrix contains feature values ​​below the identifiable threshold, the information increment corresponding to each candidate scan angle is calculated based on the predicted observation Jacobian matrix and the predicted measurement weight matrix corresponding to the candidate scan angle. The candidate scan angle with the largest information increment is selected, and the axial scanning interval, cross-sectional direction angle range, scanning device position, and scanning device orientation corresponding to the candidate scan angle are output.

[0017] Preferably, the scanned point cloud is converted to the original unified engineering coordinate system. A local update domain is determined based on the axial scanning interval and cross-sectional orientation angle range covered by the scanned point cloud. Only the measurement confidence weight, generation source label, material layer axial termination boundary, local cross-sectional deformation matrix, and reference axis control points within the local update domain are updated. The original point cloud version, scanned point cloud version, reference axis parameters, material layer standard cross-sectional profile, axial termination boundary, generation source label, counterfactual reconstruction residual, and original point index are written into the same data record.

[0018] Beneficial effects

[0019] This invention provides a robust reference axis modeling and point cloud segmentation method for cable joints. It has the following beneficial effects:

[0020] 1. The present invention uses a unified structural inversion reference axis based on the unstripped parent body. The reference axis does not switch with the change of the exposed material layer in the axial section, forming a unified coordinate reference for the core length, material layer stripping length, fracture inclination and cross-sectional size.

[0021] 2. This invention transforms point cloud segmentation into point generation source identification. Both the longitudinal surface and the cut surface of the material layer have corresponding solid structure sources, and the material layer category does not depend on a single color threshold or a single local curvature feature.

[0022] 3. This invention determines whether the inner layer material has the conditions for exposure based on the termination state of the outer layer material, distinguishes between actual material removal and scanning obstruction, and avoids directly recording scanning blind spots as construction defects.

[0023] 4. The present invention sets up counterfactual hidden verification of stable axial segments, so that the reference axis and material layer contour can be independently verified by other axial segments, thereby reducing the overfitting of local point clouds to the parent model.

[0024] 5. This invention determines the direction of insufficient observation based on the information matrix of model parameters, and determines the scanning attitude based on the information increment of candidate scanning angles, so that the scanning data can supplement the specific uncertain parameters in the reference axis, cross-sectional deformation and shearing boundary. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the overall process of the robust reference axis modeling and point cloud segmentation method for cable joints according to the present invention.

[0026] Figure 2 This is a diagram showing the composition of the virtual unstripped cable parent body according to the present invention.

[0027] Figure 3 This is a diagram showing the relationship between the reference axis, the local section base, and the local section deformation matrix of this invention.

[0028] Figure 4 This is a diagram showing the relationship between the longitudinal surface of the material layer, the cut surface, and the source label of the present invention.

[0029] Figure 5 This is a diagram showing the relationship between the axial termination boundary of the material layer of the present invention and the axial-direction angle cylindrical domain;

[0030] Figure 6 This is a flowchart illustrating the point cloud generation and source tag update process of this invention.

[0031] Figure 7 This is a flowchart of the counterfactual hiding verification process for the stable axial segment of the present invention;

[0032] Figure 8 This is a flowchart illustrating the actual peeling, scanning occlusion, and undeterminable state discrimination process of this invention.

[0033] Figure 9 This is a flowchart of the model identifiability analysis and directional scan of the present invention;

[0034] Figure 10 This is a diagram showing the relationship between the point cloud, parent parameters, segmentation labels, and engineering data evidence chain of this invention.

[0035] Figure 11 This is a screenshot of the main interface of the system of the present invention;

[0036] Figure 12 This is a screenshot of the system functions of the present invention. Detailed Implementation

[0037] 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. Specific Implementation Example 1:

[0039] like Figures 1 to 12 As shown, a robust reference axis modeling and point cloud segmentation method for cable joints treats the current point cloud of the cable joint as the observation result formed after a complete multi-layer cable has been stripped layer by layer. Based on the exposed conductor, insulation layer, semi-conductive layer, shielding layer, and outer sheath surfaces in different axial sections, a virtual unstripped cable matrix is ​​reconstructed in reverse. The virtual unstripped cable matrix consists of a reference axis in the form of a spatial curve, a local cross-sectional deformation matrix that varies along the reference axis, standard cross-sectional profiles of nested material layers from the inside out, and axial termination boundaries of the material layers that vary with the cross-sectional direction angle.

[0040] Each point in the cable joint point cloud is interpreted as being generated from the longitudinal surface of the material layer, the cut surface of the material layer, or an abnormal state outside the parent cable in the virtual unstripped cable matrix. The system determines the point's generation source label based on the normalized distance of the measurement error from the point to each generated surface, the difference in material properties, and the material layer retention state. Under the constraints of the material layer nesting relationship and the stripping sequence, the system jointly updates the reference axis, the local cross-sectional deformation matrix, the material layer contour, the material layer termination boundary, and the point label.

[0041] To prevent local surface overfitting, this invention sequentially sets stable axial segments of the same material layer as hidden verification segments. After excluding the point cloud of this hidden segment, it reconstructs the hidden segment based on other axial segments. Only parent parameters that meet the hidden segment reconstruction error threshold are included in the final result. This process ensures that the reference axis is constrained by multiple axial segments and multiple material layers, rather than being determined by a local arc or a single complete cross-section.

[0042] The method of the present invention is executed in the following order: point cloud acquisition and coordinate unification, measurement credibility modeling, initial reference axis generation, non-torsion curve coordinate establishment, virtual unstripped cable matrix initialization, material layer retention state determination, generation source label update, matrix parameter joint inversion, counterfactual hiding verification, identifiability analysis, directional scan, and result output.

[0043] In the embodiment described, the material layer numbers increase from the inside out. The cable structure template determines the actual number and names of the material layers involved in the modeling, while the joint stripping process template determines the stripping direction, material layer termination sequence, minimum distance between adjacent termination boundaries, size determination interval, and effective scanning area.

[0044] The original point cloud refers to the set of spatially discrete points acquired by a 3D scanning device from the surface of a cable joint. Each point records its 3D coordinates, color, reflection intensity, normal vector, scanning incident direction, device confidence level, and scan frame number. The color consists of red, green, and blue channels; the reflection intensity is the normalized intensity output by the scanning device; and the normal vector is determined by the unit eigenvector corresponding to the minimum eigenvalue of the spatial neighborhood covariance matrix of the point.

[0045] The virtual unstripped cable matrix refers to a complete multilayer cable geometric model built in reverse based on the visible material surfaces and cut surfaces after stripping. The model retains the structural relationship of the stripped material layers before construction, but does not write the counterfactual hidden layer into the actual observation point cloud.

[0046] The longitudinal surface of a material layer refers to the outer peripheral surface of the material layer extending along the cable axis. The cut surface of a material layer refers to the transition surface connecting the outer peripheral contour of the material layer and the inner contour of the material layer at the axial termination position. The generation source label indicates that the points in the point cloud originate from a longitudinal surface of a material layer, a cut surface of a material layer, or a non-parent abnormal state.

[0047] A stable axial segment refers to a point cloud segment on the longitudinal surface of a material layer that is far from the axial termination boundary of the material layer, has a continuous axial length, an orientation angle coverage that reaches the coverage threshold, and a residual from the point to the corresponding parent surface that reaches the residual threshold. Counterfactual hiding verification involves excluding a stable axial segment from the parameter estimation data, reconstructing the excluded segment based on the remaining segments, and evaluating the parent parameters based on the residual between the reconstructed location and the actual point cloud.

[0048] The original point cloud of the cable joint is acquired by a structured light scanning device, a laser 3D scanning device, or an industrial vision device with depth output. This implementation uses multi-view scanning, with a surface overlap rate of no less than 30% between scan frames, and each scan frame covering 20% ​​to 80% of the joint's axial length. The nominal point spacing of the scanning device is no greater than one-quarter of the thickness of the thinnest material layer being measured, and the nominal length error of the device is no greater than one-third of the allowable dimensional error for the project.

[0049] The scanning equipment calibration data includes equipment intrinsic parameters, scale correction coefficients, rotation matrix, and translation vector for each scan frame. The cable structure template is derived from the cable product structure data, recording material layer numbers, material names, nominal radius, nominal thickness, lower thickness limit, and upper thickness limit. The joint stripping process template is derived from the corresponding joint construction process document, recording the stripping direction, material layer termination sequence, minimum axial spacing between adjacent termination boundaries, cut protection distance, and engineering dimension thresholds.

[0050] The system establishes a data directory according to the project number, and saves the original scan files, equipment calibration files, cable structure templates, and joint stripping process templates separately. Point cloud preprocessing, parent model inversion, and quantity calculation all read data under the same project number, and the input file hash value and template version number are written into the output record.

[0051] Unifying point cloud coordinates from multiple perspectives: (1), Indicates the first The first scan frame The three-dimensional coordinates of the original point within the local coordinate system of the scanning device; This represents the three-dimensional coordinates after transformation to a unified engineering coordinate system; Indicates the first A three-dimensional rotation matrix for each scan frame; Indicates the first The three-dimensional translation vector of each scan frame; Indicates the point number within the scan frame; Indicates the scan frame number.

[0052] The system first performs an initial coordinate transformation based on the device pose, and then establishes corresponding point pairs within the overlapping area of ​​adjacent scan frames. Corresponding points are selected from regions that are axially continuous, have local roughness below a roughness threshold, and are more than the boundary protection distance from the point cloud boundary. Cut points, conductor stray points, and isolated points are not included in the initial set of corresponding point pairs.

[0053] (2), This indicates the registration error between the point cloud to be registered and the reference point cloud. Represents the set of corresponding point pairs; Indicates the first point in the point cloud to be registered. One point; Indicates the first point in the reference point cloud One corresponding point; Indicates reference point The unit normal vector at that location; This represents the rotation matrix to be optimized; This represents the translation vector to be optimized; Represents corresponding point pairs Registration weights; superscript This represents the transpose of a vector.

[0054] The Gauss-Newton iterative solution formula (2) is used. In each iteration, the nearest neighbor points are re-established and points whose distance to the plane exceeds three times the median distance are deleted. The rotation increment is less than... Registration ends when the radian and translation increment are less than 10% of the nominal point spacing of the scan.

[0055] Point-level measurement confidence weights:

[0056] (3), Indicates the first The overall measurement confidence weight of each point; Indicates the local point density weight; Indicates the neighborhood normal consistency weight; Indicates the weight of the scanning incident direction; Indicates the weight of overlapping residuals across multiple frames; This indicates the confidence weight of the data collected by the device.

[0057] The local point density weight is obtained by truncating the ratio of the actual number of points to the number of reference points within the radius neighborhood to a value between zero and one. The neighborhood normal consistency weight is determined by the weighted variance of the angle between the neighborhood normal vector and the center point normal vector. The scanning incident direction weight is determined by the absolute value of the cosine of the angle between the point normal vector and the observation direction from the scanning device to that point. The multi-frame coincidence residual weight is determined by the residuals of corresponding points on the same entity surface in different scanning frames. The device acquisition confidence weight is jointly determined by the device confidence level and the reflection intensity saturation label.

[0058] Low-confidence points are not removed from the original point cloud. The calculation integrates the distance cost from the entry point to the generated surface, material layer profile updates, reference axis updates, and counterfactual reconstruction errors, based on the comprehensive measurement confidence weight. Original point indices are retained for conductor strands, thin cuts, and material residue points.

[0059] Initial reference axis and torsion-free curve coordinate system: The system establishes an initial direction based on the axial extensibility of the point cloud. The point cloud is divided into overlapping slices along the initial direction, with each slice having a thickness of 1% to 5% of the nominal outer diameter of the cable, and the center-to-center distance between adjacent slices being half the slice thickness. Each slice uses a coordinate median weighted by comprehensive measurement reliability weights, and cubic B-splines connect all valid slice centers.

[0060] (4), Indicates the initial reference axis in the arc length parameter The three-dimensional coordinates of the location; This indicates the number of control points for the cubic B-spline; Indicates the first A cubic B-spline basis function; Indicates the first One initial three-dimensional control point; Indicates the control point number; superscript This represents the initial state.

[0061] (5), Indicates the reference axis in the arc length parameter The unit tangent vector at that point; Indicates the reference axis relative to the arc length parameter The first derivative; This represents the Euclidean norm.

[0062] At the initial position of the reference axis, a first cross-sectional unit basis vector orthogonal to the unit tangent vector is selected, and the second cross-sectional unit basis vector is determined by the cross product of the unit tangent vector and the first cross-sectional unit basis vector. When propagating the cross-sectional basis vectors along the reference axis, the minimum rotation between adjacent unit tangent vectors is calculated, and the minimum rotation is applied to the two cross-sectional basis vectors at the previous axial position to form a torsion-free cross-sectional coordinate system.

[0063] (6), Indicates the first The corresponding arc length parameter of each point on the reference axis; Represents the independent variable that minimizes the objective function; Indicates the first The three-dimensional coordinates of each point; Indicates the reference axis; This indicates the reference axis arc length parameter.

[0064] (7), Indicates the first Two-dimensional cross-sectional coordinates of a point; Represents the local cross-section basis matrix at the corresponding location; This represents the transpose of the local section basis matrix; Indicates the first The three-dimensional coordinates of each point; This indicates the corresponding reference axis point.

[0065] The local section basis matrix is ​​composed of the unit basis vectors of the first and second sections, arranged column-wise. The first and second components of the two-dimensional section coordinates correspond to the two section basis directions, respectively. The section orientation angle of a point is calculated using the two-parameter arctangent function, and the radial distance of a point is calculated using the Euclidean norm of the two-dimensional section coordinates.

[0066] Virtual unstripped cable mother:

[0067] (8), Represents arc length parameter The two-dimensional local cross-sectional deformation matrix at the location; and These represent the expansion and contraction parameters in the two cross-sectional base directions, respectively; This represents the coupling deformation parameter between the two cross-sectional base directions; This indicates that the matrix is ​​a positive definite matrix.

[0068] (9), Represents a two-dimensional lower triangular matrix with positive diagonal elements; Represents the local cross-sectional deformation matrix; superscript This indicates the matrix transpose.

[0069] The system uses three independent elements of a lower triangular matrix as optimization variables, with two diagonal elements parameterized exponentially to ensure that the local cross-sectional deformation matrix remains positive definite throughout the entire iteration process. The local cross-sectional deformation matrix is ​​constrained by a second-order difference along the axial direction. The actual flattening of the cross-section is expressed by this matrix, while lateral movement of the reference axis is not written into this matrix.

[0070] (10) Indicates the first Each cross-sectional orientation angle; Represents pi; Indicates the number of nodes in the direction angle; Indicates the direction angle node number.

[0071] (11), Indicates the first The material layer in the first Two-dimensional standard cross-sectional profile points at each directional angle; Indicates the standard radial distance at the corresponding direction angle; This indicates the material layer numbering arranged from the inside out; and They represent the direction angles respectively. The cosine and sine values.

[0072] (12), Indicates the outermost number The material layer in the first Radial distance at each directional angle; Indicates the inner side of the first The radial distance between the material layers at the same angular direction; Indicates the first The material layer and the first The minimum radial spacing between material layers.

[0073] (13) Indicates the first The material layer in the first Two-dimensional standard cross-sectional profile points at each directional angle; Indicates the number of nodes in the direction angle; This represents a two-dimensional zero vector.

[0074] (14) Indicates the effective arc length of the reference axis; Represents the local cross-sectional deformation matrix; Represents the determinant of a matrix; Represents the natural logarithm; This represents the arc-length integral element.

[0075] Formula (13) restricts the centroid of the standard cross-section profile of each material layer to be located on the reference axis, and Formula (14) restricts the average area scaling of the local cross-section deformation matrix along the reference axis. The above two constraints eliminate the redundant expressions between the translation of the reference axis and the eccentricity of the material layer, and the redundant expressions between the standard scale of the material layer and the scale of the local cross-section, respectively.

[0076] (15) Indicates the first Each material layer in arc length parameter and cross-sectional orientation angle Three-dimensional longitudinal surface point at the location; Indicates the reference axis; Represents the basis matrix of the local section; Represents the local cross-sectional deformation matrix; Indicates the first The standard cross-sectional profile of each material layer.

[0077] Direction-related axial termination boundary and cut surface: (16) Indicates the first Each material layer at the cross-sectional direction angle The axial termination position at; Indicates the reference axis arc length parameter; Indicates the material layer number; Indicates the cross-sectional orientation angle.

[0078] (17) Indicates the first Three-dimensional points on the surface of a material layer cut; Indicates the first Each material layer at the orientation angle The axial termination position at; Indicates from the inner side The material layer profile to the first The radial interpolation parameters of each material layer profile, and ; Indicates the outline of the inner material layer; Indicates the first The material layer outline; the meanings of the remaining symbols are the same as those in formula (15).

[0079] When the cut has a gentle slope, the axial termination boundary is extended to a width parameter, and the cut surface is distributed along the axial width range. The width parameter is determined by the axial distribution of the candidate cut points. When the cut width is less than two axial nodes, it is treated as a single termination position.

[0080] Material layer retention state and theoretically exposed material layer:

[0081] (18) Indicates the first Each material layer in arc length parameter and cross-sectional orientation angle The status of the location is preserved; This indicates an indicator function; the value is taken when the condition within the parentheses is true. If the condition in parentheses is not met, then take... ; This represents the peeling propulsion direction coefficient, taken when the positive direction of the reference axis coincides with the peeling propulsion direction. When the direction is opposite, take ; This indicates the axial termination boundary of the material layer.

[0082] (19) Represents arc length parameter and cross-sectional orientation angle The theoretically outermost layer of retained material is numbered; This indicates that the largest number is selected from the material layer numbers that meet the conditions; Indicates the first The retention state of each material layer.

[0083] The system establishes a set of candidate generated surfaces based on theoretically exposed material layers. Where the outer material layer is still preserved, the longitudinal surface of the inner material layer does not enter the candidate generated surface set; where the outer material layer has terminated, the longitudinal surface of the inner material layer enters the candidate generated surface set. Points near the axial termination boundary of the material layer simultaneously possess both longitudinal surface and notch surface candidates.

[0084] Source tags generated by the point: (20) Indicates the first Points to candidate generation surface The squared Mahalanobis distance; Indicates the first The three-dimensional coordinates of each point; Indicates the first Points on the candidate generation surface The nearest projection point on; Indicates the first The three-dimensional measurement error covariance matrix of points; superscript Indicates matrix inversion; superscript This represents the transpose of a vector.

[0085] The nearest projection of a point onto the longitudinal surface is obtained by alternately updating the axial parameters and the cross-sectional orientation angle. First, the local cross-section is determined using the point's current axial parameters, then a one-dimensional interpolation search is performed between adjacent orientation angle nodes; finally, a one-dimensional Newton search is performed along the reference axis with the orientation angle fixed. The nearest projection of a point onto the cut surface is then subjected to a two-dimensional local search using the cross-sectional orientation angle and radial interpolation parameters.

[0086] (twenty one), Indicates the first The source tags for each point; Indicates the first A set of candidate generated surfaces for each point; This represents a candidate generated surface that minimizes the overall cost. Indicates the first The color, reflection intensity, normal vector, and local roughness of each point are related to the candidate generated surface. The cost of differences between corresponding material properties; Indicates candidate generated surfaces The cost of conflict between the retention state of the material layers and the stripping sequence; This indicates the weight of the cost of material property differences; This represents the weight of the conflict cost for preserving the state.

[0087] The material property difference cost is calculated using the mean vector and covariance matrix obtained from the statistics of stable segments of each material layer. In the first iteration, when there are no stable segments, the weight of the material property difference cost is set to zero, and the label is updated only based on geometric distance and preserved state. After the start of the second iteration, the material property statistics are updated based on high confidence points. When the candidate generated surface does not conform to the theoretical exposed material layer relationship, the preserved state conflict cost is taken as a constant greater than the upper bound of the cost of all legal candidates.

[0088] Anomalies are assigned an independent entry cost. When the Mahalanobis distance from a point to all legally generated surfaces exceeds the anomaly threshold, the point enters a non-parent anomaly state. Anomalies are further classified into conductor strands, burrs, material attachments, external clamps, and isolated noise points based on their connectivity with the conductor body, axial continuity, local radial deviation, and neighborhood normal distribution.

[0089] Joint inversion of parent parameters: (twenty two), This represents the set of parent parameters to be determined. Indicates the reference axis; This represents the local cross-sectional deformation matrix distributed along the reference axis; to Indicates the first The material layer up to the first Standard cross-sectional profile of each material layer; to Indicates the first The material layer up to the first Axial termination boundary of each material layer; to Indicates the first Point 1 to the 2nd The source tags for each point; Indicates the number of material layers; This indicates the number of points in the point cloud.

[0090] (twenty three), Represents the optimal set of parent parameters; Indicates the first The overall measurement confidence weight of each point; Indicates the first The Mahalanobis distance from each point to the generated surface corresponding to its source label; The threshold parameter is Robust loss function; Represents the weights of physical structure constraints; This represents the physical structural constraints consisting of reference axis continuity, local cross-sectional deformation continuity, radial nesting of material layers, peeling sequence, and parameter uniqueness. Indicates the counterfactual verification weight; This indicates a counterfactual reconstruction of the residual.

[0091] (twenty four), Indicates the input residual The corresponding robust loss value; Represents the standardized distance residual; Represents the absolute value of the residual; This represents the residual threshold after the quadratic penalty is converted to a linear penalty.

[0092] The physical structure constraints include the square integral of the second derivative of the reference axis, the square integral of the third derivative of the reference axis, the square sum of the second difference of the local cross-section deformation matrix along the axis, the square sum of the second difference of the material layer profile angle, the penalty for adjacent material layers violating the minimum radial spacing, the penalty for the axial termination boundary violating the peeling order, the profile centroid constraint corresponding to formula (13), and the average area scaling constraint corresponding to formula (14).

[0093] The joint inversion employs three levels of calculation: coarse resolution, medium resolution, and fine resolution. The coarse resolution uses representative points with voxel sizes two to four times the original average point spacing, and thirty-six angular nodes, to determine the number of material layers and the approximate range of the termination boundary. The medium resolution uses representative points with voxel sizes one to two times the original average point spacing, and seventy-two angular nodes, to update the reference axis, local section deformation matrix, and material layer profile. The fine resolution restores the original point index, with one hundred and twenty or one hundred and forty-four angular nodes, and updates the cut boundary and local outliers.

[0094] When generating source labels and fixing the reference axis, each longitudinal surface point of the material layer is back-mapped to the standard section through the inverse matrix of the current local section deformation matrix. Each orientation angle node collects the back-mapped radial distance within adjacent angular sectors, calculates the weighted median with the comprehensive measurement confidence weight as the weight, and performs radial nested projection according to formula (12). Orient angle nodes without direct observation are jointly determined by the counterfactual transport values ​​of other axially stable segments of the same material layer and the second-order difference constraint of the profile angle.

[0095] With the standard profile of the material layer fixed, the lower triangular matrix parameters of each axial window are updated using all longitudinal surface points of the stable material layer. The length of the axial window is taken as half to twice the nominal outer diameter of the cable, and adjacent windows overlap by 50%. The updated local cross-sectional deformation matrix is ​​continuously interpolated along the reference axis using cubic splines.

[0096] Solving for the axial termination boundary of the material layer: (25), Represents the set of optimal axial termination boundaries; Represents the set of axial termination boundaries to be determined; Indicates the first The material layer in the first Select axial position at each directional angle Local cost at the termination position; Indicates the number of material layers; Indicates the number of nodes in the direction angle; Indicates the weight of continuity around the boundary circle; This indicates the difference in axial boundary position at adjacent direction angles; Indicates the weight of the peeling order constraint; This indicates a penalty when the axial termination boundary violates the material layer peeling sequence.

[0097] The local boundary cost consists of four components: termination evidence (the number of longitudinal surface points of the corresponding material layer decreases along the axial direction), incision evidence (the appearance of candidate points on the incision surface), exposure evidence (the increase in the number of longitudinal surface points of the inner material layer), and attribute evidence (the statistical change in material properties). Each piece of evidence is normalized to zero to one within the axial window. The stronger the termination evidence, incision evidence, and inner layer exposure evidence, the lower the local boundary cost.

[0098] In the discrete implementation, all candidate axial positions on each orientation angle node are treated as graph nodes. Continuous connections are established between adjacent orientation angle nodes within the same material layer, and peeling sequence connections are established between the same orientation angle nodes for different material layers. Minimum cost flow is used to solve for multiple non-intersecting material layer termination boundaries that satisfy the minimum axial spacing. The orientation angle nodes are connected end-to-end to form a circumferential closed boundary.

[0099] Reference axis update:

[0100] (26) This indicates the objective function for updating the reference axis; This represents the set of stable points that participate in the reference axis update; Indicates the first The overall measurement confidence weight of each point; Indicates the first The squared Mahalanobis distance from each point to its currently generated surface; Indicates the effective arc length of the reference axis; Indicates the reference axis relative to the arc length parameter The second derivative; Indicates the reference axis relative to the arc length parameter The third derivative; Indicates the curvature constraint weight; Indicates the curvature variation constraint weight; This represents the arc-length integral element.

[0101] The stable point set excludes points that are less than the axial termination boundary of the material layer by the cut protection distance, scan-obstructed points, counterfactual inference points, non-parent anomalies, and axial window points with orientation angle coverage below the coverage threshold. The Gaussian-Newton method is used to update the B-spline control points, limiting the movement of a single control point to half the current voxel size in each iteration. After updating the reference axis, the arc length parameters, section orientation angles, and most recently generated surface projections of the points are recalculated.

[0102] Counterfactual hiding verification:

[0103] (27) Indicates the first The counterfactual reconstruction error of a hidden verification fragment; Indicates the first A set of points in a hidden verification fragment; Indicates the first The overall measurement confidence weight of each point; Indicates the first One practical point; Indicates based on the exclusion of the first The parent parameter prediction obtained after the hidden verification fragment is the first... Location of each point; Indicates exclusion of the first The set of parent parameters obtained after hiding the verification fragment.

[0104] (28) This indicates a comprehensive counterfactual reconstruction of the residual; Indicates the number of hidden verification fragments; This indicates that the verification segment number is hidden. Indicates the first Counterfactual reconstruction error of a hidden verification fragment.

[0105] Stable segments within the same material layer are divided according to their axial position. The axial length of each stable segment is not less than half of the nominal outer diameter of the cable, the directional angle coverage is not less than 50%, and the weighted root mean square residual from the point to the longitudinal surface of the material layer is not greater than twice the nominal length error of the equipment. Hidden verification is performed sequentially for each stable segment. Hidden segments do not participate in the updating of the reference axis, material layer profile, and local cross-sectional deformation matrix.

[0106] Hidden segment prediction uses the standard profile, reference axis, and interpolated values ​​of the local cross-sectional deformation matrix at the adjacent axial positions of the hidden segment, all determined by other segments in the same layer. When the counterfactual reconstruction error of the hidden segment exceeds the counterfactual threshold, the system sequentially checks the reference axis control point, the local cross-sectional deformation matrix, and the material layer number, and performs damped updates on the parameter that contributes the most to the error.

[0107] Alternating solution and convergence criteria:

[0108] (29) Indicates the first The average change of the reference axis in each iteration; Indicates the effective arc length of the reference axis; Indicates the first The reference axis for round iteration; Indicates the first The reference axis for round iteration; Indicates the iteration round.

[0109] (30) Indicates the first The average change of the axial termination boundary in each iteration; Indicates the number of material layers; Indicates the number of nodes in the direction angle; Indicates the first The first iteration obtained The first material layer Axial termination boundary of one direction angle; Indicates the first The axial termination boundary corresponding to the round iteration.

[0110] (31), Indicates the first The rate of change of the source label in each iteration; Indicates the number of points in the point cloud; Indicates the first In the first iteration The source tags for each point; Indicates the first In the first iteration The source tags for each point; This indicates an indicator function.

[0111] Each round of joint iteration sequentially performs source label updates, material layer standard profile updates, local cross-sectional deformation matrix updates, material layer axial termination boundary updates, reference axis updates, and counterfactual hiding verification. The joint iteration at the current resolution level ends when the average change in the reference axis is less than 20% of the current voxel size, the average change in the axial termination boundary is less than half the axial node spacing, and the change rate of the source labels is less than 0.5%. After meeting these conditions in the fine-resolution stage, the final parent parameters are output.

[0112] The system first determines the theoretical exposed material layer at each axial node and directional angular node according to formulas (18) and (19), and then establishes a line of sight from the center of the scanning device to the theoretical surface point based on the scanning device pose. When the line of sight intersects with the observed outer surface, the corresponding position is marked as a scan occlusion. When the lines of sight do not intersect but the scanning incident angle exceeds the effective incident angle calibrated by the device, it is also marked as a scan occlusion.

[0113] When the axial termination boundary of the outer material layer is located before the current axial position, and the inner material layer has actual point cloud support from three consecutive axial nodes and three consecutive directional angular nodes, the inner material layer is marked as the actual material removal area. When the termination state of the outer material layer lacks cutting evidence and the scan occlusion relationship is not established, the corresponding area is marked as an undeterminable area.

[0114] Undeterminable areas are not included in the calculation of engineering dimensional compliance. In the 3D display, the system uses different fill textures to represent the directly observed area, the actual stripped area, the scan occluded area, the counterfactual inference area, and the undeterminable area, and writes the corresponding status code in the data record.

[0115] Model discernibility and directional scan:

[0116] (32), Indicates candidate fill-in view Information increment; An information matrix representing the current parent parameters; Indicates candidate fill-in view The corresponding predicted observation Jacobian matrix; Indicates candidate fill-in view The corresponding prediction measurement weight matrix; Represents the determinant of a matrix; Represents the natural logarithm; superscript This indicates the matrix transpose.

[0117] (33), Indicates the selected scan viewpoint; Represents the set of candidate supplementary scan views; Indicates the candidate fill-in viewpoint; This represents a candidate perspective that maximizes the information increment. Indicates candidate fill-in view The information increment.

[0118] The information matrix is ​​formed by the Gaussian-Newton approximation matrix of the jointly inverted target at the convergence point. When the minimum eigenvalue of the information matrix is ​​lower than the discriminable threshold, the system generates a set of candidate supplementary scanning perspectives. The candidate supplementary scanning perspectives are discretized by the allowed equipment position, scanning distance range, equipment field of view, and safety isolation area on site.

[0119] For each candidate supplementary scanning viewpoint, the system performs line-of-sight projection on the virtual matrix surface, retaining unobstructed predicted observation points with incident angles within the effective range of the device, and calculating the observation Jacobian matrix and measurement weight matrix based on the predicted observation points. The supplementary scanning viewpoint corresponding to formula (33) is selected, and the axial scanning interval, orientation angle range, device position, and device orientation are output.

[0120] After importing the supplementary scan data, perform the same coordinate transformation as in formula (1), and establish a local update domain based on the axial interval and azimuth range covered by the supplementary scan point cloud. The local update domain is extended axially by a reference axis control point spacing and azimuthally by two azimuth nodes. Only the point labels, axial termination boundary, local cross-sectional deformation matrix, and reference axis control points within the local update domain are updated.

[0121] Quantity calculation: (34), Indicates the first The first material layer and the first The material layer in the first Axial boundary distance at each directional angle; Indicates the first The material layer in the first Axial termination boundary at each directional angle; Indicates the first The axial termination boundary of each material layer at the same directional angle; and Indicates the material layer number; Indicates the direction angle node number.

[0122] (35), Indicates the first The circumferential consistency index of the fracture surface of each material layer; Indicates the number of nodes in the direction angle; Indicates the first The material layer in the first Axial termination boundary at each directional angle; Indicates the first The arithmetic mean of the axial termination boundaries of all directions of each material layer.

[0123] (36) Indicates the ellipticity of the cross section; This represents the maximum cross-sectional diameter calculated based on the standard cross-sectional profile of the material layer and the local cross-sectional deformation matrix. This indicates the corresponding minimum cross-sectional diameter.

[0124] The system calculates the peeling length of adjacent material layers across all effective directional angles, outputting the minimum peeling length, maximum peeling length, arithmetic mean peeling length, and peeling length range. The fracture circumference consistency index is directly calculated using the direction-dependent axial termination boundary. Insulation diameter and ellipticity are calculated within the stable axial segment of the insulation; scan-blocked regions and counterfactual inference regions are excluded from the final diameter extreme value.

[0125] Parameter settings and software implementation: The average point spacing of the original point cloud is denoted as... The coarse resolution voxel size is set to... Medium resolution voxel size set to Fine resolution uses the original point cloud. Axial node spacing is set to... to The spacing between the reference axis B spline control points is set to be less than half the axial width of the shortest cut to be identified.

[0126] The number of directional nodes is set to 36 in the coarse resolution stage, 72 in the medium resolution stage, and 120 in the fine resolution stage. The cut protection distance is set to 5% to 20% of the cable's nominal outer diameter. The minimum stable axial segment length is set to half the cable's nominal outer diameter. The abnormal distance threshold is set to three to five times the equipment's nominal length error.

[0127] Point cloud neighborhood retrieval, point-to-surface projection, and sparse nonlinear least squares are implemented in C++, while project management, template management, and report generation are implemented in Java. 3D display is achieved using a graphics engine that supports point clouds, curves, and surface objects. The parent parameters are stored using a sparse block matrix, and independent parameter blocks are established for reference axis control points, axial node local deformation parameters, material layer contour nodes, and material layer boundary nodes.

[0128] Each point record includes the original point number, scan frame number, uniform engineering coordinates, measurement confidence weight, generation source label, material layer number, surface type, Mahalanobis distance, evidence status, and model version. Each material layer record includes the standard profile, axial termination boundary, stable fragment index, and counterfactual reconstruction error. Specific Implementation Example 2:

[0130] This embodiment processes a cable joint consisting of a complete outer sheath section, an exposed semi-conductive layer section, an exposed insulation section, and a bare conductor section. Point clouds are acquired using four scanning perspectives, located sequentially from the front, left, right, and slightly above the back of the joint. The scanning equipment has a nominal point spacing of 0.2 mm and a nominal length error of 0.05 mm. The cable's nominal outer diameter is 65 mm, and the effective scanning length of the joint is 480 mm.

[0131] After unifying the point cloud coordinates, it contains 3.2 million original points. The coarse resolution voxel size is set to 0.6 mm, the medium resolution voxel size is set to 0.3 mm, the axial node spacing is set to 1.2 mm, and the initial spacing of the reference axis control points is set to 40 mm. The cable structure template includes four material layers: conductor, insulation layer, semi-conductive layer, and outer sheath. The number of directional angular nodes is set to 120 in the fine resolution stage.

[0132] The system establishes an initial reference axis using a robust principal direction and the center of overlapping slices. The first round of label updates uses only the Mahalanobis distance from the point to the generated surface and the material layer retention status. After completion, material property statistics are established from the high-confidence longitudinal surface points of each material layer. Complete outer sheath segments are assigned to the outer sheath longitudinal surface, exposed semiconducting layer segments to the semiconducting layer longitudinal surface, exposed insulation segments to the insulation longitudinal surface, and conductor body segments to the conductor longitudinal surface. Points deviating from the conductor body contour but continuous with the conductor ends are assigned to the conductor strand state.

[0133] The system solves for the outer sheath termination boundary based on evidence of the longitudinal surface termination of the outer sheath, evidence of the outer sheath cut, and evidence of the beginning of exposure of the semiconductive layer; the system solves for the termination boundary of the semiconductive layer and the termination boundary of the insulating layer in the same way. Points within five millimeters on both sides of the boundary are not included in the set of stable points of the reference axis. After completing the update of the material layer profile and the local cross-sectional deformation matrix, the reference axis is updated using formula (26).

[0134] The exposed insulation section is divided into three stable axial segments, and counterfactual hiding verification is performed sequentially. When hiding an insulation segment, the segment is reconstructed using the standard profile, reference axis, and local cross-sectional deformation matrix of adjacent axial positions determined by the other insulation segments. After the counterfactual reconstruction error of all three hidden segments is less than 0.15 mm, the current parent parameters are retained.

[0135] In the point cloud missing area of ​​the complete outer sheath section, the outer sheath is preserved in state 1, and the line of sight is blocked by the cable body. The system marks this area as a scan-blocked area. In the partially missing area of ​​the insulation exposed section, the axial termination boundaries of both the outer sheath and the semiconductive layer are located before the current axial position, and there are direct observation points for the adjacent direction angles of the insulation. The system marks this area as an insulation counterfactual inference area.

[0136] The system outputs a unified reference axis, longitudinal surface point clouds of four material layers, cut surface point clouds of three material layers, conductor strand point clouds, scan occlusion area, counterfactual inference area and axial termination boundary related to each material layer direction, and calculates the stripping length, fracture circumference consistency and insulation ellipticity according to formulas (34) to (36). Specific Implementation Example 3:

[0138] This embodiment addresses cable joints with obliquely cut outer sheaths and insufficient point cloud on the back side. The initial scan employs three perspectives: front, left, and right. The outer sheath retains a longer length in the front view than in the back view, while the back view lacks continuous cut points.

[0139] The system establishes the axial termination boundary of the outer sheath based on the longitudinal surface termination point, cut surface point, and semiconductive layer exposure point of the front and side outer sheaths. The axial interval width of the boundary of the back-side corner node exceeds half of the allowable error of the project boundary, and the minimum eigenvalue of the parent parameter information matrix is ​​lower than the identifiable threshold.

[0140] The system generates three candidate supplementary scanning angles—horizontal, obliquely upward, and obliquely downward—within the reachable field area, and calculates the predicted observation Jacobian matrix, predicted measurement weight matrix, and information increment for each angle. The obliquely upward angle has the largest information increment, and the system outputs the scanning equipment position, equipment orientation, axial scanning interval, and azimuth angle range corresponding to this angle.

[0141] After the point cloud is transformed to a unified engineering coordinate system, the system extends the reference axis control point spacing on both sides of the scanned coverage area to form a local update domain. Only the outer sheath termination boundary, local cross-sectional deformation matrix, and adjacent reference axis control points within the back orientation angle range are updated. The parent parameters of the front and side sides, which have been verified by counterfactual hiding, remain unchanged.

[0142] After the local update is completed, a closed three-dimensional cut boundary of the outer sheath is formed, and the initial scan point, the supplementary scan point, and the counterfactual inference boundary are recorded as independent evidence states.

[0143] The project acceptance data record includes the original point cloud version, the supplementary point cloud version, the input file hash value, the scanning equipment calibration version, the cable structure template version, the joint stripping process template version, the reference axis control points, the local cross-sectional deformation matrix, the standard cross-sectional profile of the material layer, the axial termination boundary of the material layer, the point generation source label, the counterfactual reconstruction error, the convergence status, the scan occlusion status, the undeterminable status, and the original point index.

[0144] Manual adjustments do not overwrite the original algorithm results. Manual adjustment records include the source labels generated before and after adjustment, the set of adjustment points, the person making the adjustment, the adjustment time, and the basis for the adjustment. Engineering measurements are associated with the corresponding material layer axial termination boundary, material layer longitudinal surface points, model version, and original point index.

[0145] The reference axis and material layer point cloud formed by the above method serve as the common data basis for subsequent calculation of exposed core length, insulation diameter measurement, cut inclination calculation, material residue identification, and digital acceptance record.

[0146] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0147] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A robust reference axis modeling and point cloud segmentation method for cable joints, characterized in that, The process includes the following steps: acquiring the original point cloud of the cable joint, the calibration parameters of the scanning equipment, the cable structure template and the joint stripping process template, converting the original points in different scanning frames to a unified engineering coordinate system, and determining the measurement confidence weight of each point; An initial reference axis is established based on the axial extension direction of the original point cloud, and a torsion-free curve coordinate system is established along the initial reference axis. A virtual unstripped cable matrix is ​​established, which includes a spatial curve-shaped reference axis to be determined, a local cross-sectional deformation matrix distributed along the reference axis to be determined, standard cross-sectional profiles of material layers nested from the inside to the outside, and axial termination boundaries of each material layer that vary with the cross-sectional direction angle. The retention state of each material layer at each axial position and cross-sectional direction angle is determined based on the axial termination boundaries, and the outermost material layer in the retention state is determined as the theoretical exposed material layer at the corresponding position. A point cloud generation surface set is established based on the theoretical exposed material layer, which includes the longitudinal surface of the material layer, the cut surface at the termination boundary of the material layer, and non-matrix abnormal states. Based on the distance from each point to each generated surface in the point cloud generation surface set, the material property differences, and the material layer retention status, the generation source label of each point is determined. Under the constraints of radial nesting relationship of material layers, material layer stripping order, and parameter uniqueness, the reference axis to be determined, the local cross-sectional deformation matrix, the standard cross-sectional profile of the material layer, the axial termination boundary, and the generation source label are jointly updated. A hidden verification segment is selected from the stable axial segments of the same material layer. The hidden verification segment is reconstructed based on the remaining stable axial segments after excluding the hidden verification segment. The virtual unstripped cable parent body is updated based on the counterfactual reconstruction residual between the reconstructed position and the actual point cloud of the hidden verification segment. The original point cloud is divided into longitudinal surface point clouds of each material layer, cut surface point clouds of each material layer, and non-parent body abnormal point clouds based on the updated generation source label. The material stripping area, scan occlusion area, and undeterminable area are distinguished based on the axial termination boundary of the outer material layer and the scanning view occlusion relationship. The reference axis, material layer segmentation point cloud, and material layer cut boundary verified by counterfactual reconstruction are output.

2. The robust reference axis modeling and point cloud segmentation method for cable joints according to claim 1, characterized in that, The measurement confidence weight is determined by the product of the point density weight, normal consistency weight, scanning incident direction weight, multi-frame overlap residual weight, and device acquisition confidence weight.

3. The method for robust reference axis modeling and point cloud segmentation of cable joints according to claim 1, characterized in that, The non-twisting curve coordinate system includes a reference axis unit tangent vector, a first cross-section unit basis vector, and a second cross-section unit basis vector. The first cross-section unit basis vector and the second cross-section unit basis vector propagate along the reference axis through the minimum rotation between adjacent reference axis unit tangent vectors. The parameter uniqueness constraint includes that the centroid of the standard cross-section profile of each material layer relative to the reference axis is zero, and the average area scaling of the local cross-section deformation matrix along the reference axis is zero.

4. The robust reference axis modeling and point cloud segmentation method for cable joints according to claim 1, characterized in that, The longitudinal surface of any target material layer is formed by adding the three-dimensional coordinates of the reference axis at the axial parameter, the local section basis matrix at the corresponding position, the local section deformation matrix, and the standard section profile of the target material layer at the section direction angle; the cut surface of the target material layer is formed by the three-dimensional coordinates of the reference axis at the axial termination boundary of the target material layer, the local section basis matrix, the local section deformation matrix, and the radial interpolation between the standard section profile of the adjacent inner material layer of the target material layer and the standard section profile of the target material layer.

5. The robust reference axis modeling and point cloud segmentation method for cable joints according to claim 1, characterized in that, The material layer retention state is determined based on the reference axis arc length parameter, the peeling advance direction, and the axial termination boundary of the material layer. At any reference axis arc length parameter and cross-sectional direction angle, the material layer that is still in the retention state and has the largest material layer number is determined as the theoretical exposed material layer, and the longitudinal surface of the material layer located inside the theoretical exposed material layer is excluded from the point cloud generated surface set at the corresponding position.

6. The robust reference axis modeling and point cloud segmentation method for cable joints according to claim 1, characterized in that, The generation source label is determined based on the sum of the squared Mahalanobis distance from the point to the candidate generation surface, the material property difference cost, and the material layer retention state conflict cost. When a candidate generated surface does not conform to the theoretical relationship of exposed material layers, the conflict cost of the material layer retention state is set as a prohibition cost.

7. The method for robust reference axis modeling and point cloud segmentation of cable joints according to claim 1, characterized in that, The joint update is performed cyclically in the following order: update the source label; update the standard cross-sectional profile and local cross-sectional deformation matrix of the material layer; update the axial termination boundary of each material layer; establish a set of stable points from the longitudinal surface points of the material layer, and update the reference axis according to the set of stable points; terminate the loop when the average change of the reference axis, the average change of the axial termination boundary, and the change rate of the source label all reach the corresponding convergence threshold.

8. The method for robust reference axis modeling and point cloud segmentation of cable joints according to claim 1, characterized in that, The hidden verification segment is composed of points in the same material layer whose axial continuous length reaches the length threshold, whose orientation angle coverage reaches the coverage threshold, and whose distance from the axial termination boundary reaches the protection distance. After excluding the hidden verification segment, the hidden verification segment is reconstructed based on the remaining stable axial segments in the same layer, the standard cross-sectional profile of adjacent material layers, the local cross-sectional deformation matrix, and the reference axis. The counterfactual reconstruction residual is determined based on the weighted average squared distance.

9. The robust reference axis modeling and point cloud segmentation method for cable joints according to claim 1, characterized in that, When the parameter information matrix of the virtual unstripped cable matrix contains feature values ​​below the identifiable threshold, the information increment corresponding to each candidate scan angle is calculated based on the predicted observation Jacobian matrix and the predicted measurement weight matrix corresponding to the candidate scan angle. The candidate scan angle with the largest information increment is selected, and the axial scanning interval, cross-sectional direction angle range, scanning device position, and scanning device orientation corresponding to the candidate scan angle are output.

10. The method for robust reference axis modeling and point cloud segmentation of cable joints according to claim 9, characterized in that, The scanned point cloud is converted to the original unified engineering coordinate system. The local update domain is determined based on the axial scanning interval and cross-sectional orientation angle range covered by the scanned point cloud. Only the generation source label, material layer axial termination boundary, local cross-sectional deformation matrix and reference axis control points within the local update domain are updated. The original point cloud version, scanned point cloud version, reference axis parameters, axial termination boundary, generation source label, reverse fact reconstruction residual and original point index are written into the same data record.

Images