A weld feature detection method in a pipeline butt joint scene
By constructing point cloud features and combining coarse and fine registration methods, the problem of high precision and automation of weld feature detection in pipeline docking scenarios is solved, realizing high-precision weld feature extraction and post-weld interlayer quality inspection, which is suitable for pipeline docking with complex structures.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING YANGOU TECH CO LTD
- Filing Date
- 2026-03-03
- Publication Date
- 2026-06-02
AI Technical Summary
Existing technologies struggle to achieve high-precision, automated weld feature detection in pipeline docking scenarios, especially in complex structures and cylindrical symmetry, where high-precision attitude estimation and weld feature extraction are difficult to achieve.
By constructing point cloud features and combining two coarse and fine registration methods, the pipe connection is performed by fitting the marker point cloud and the cylindrical model, and point cloud slice analysis is performed to extract the geometric feature parameters of the weld.
It achieves high-precision weld feature extraction, improves detection accuracy and robustness, supports online detection of interlayer quality after welding, and is applicable to pipe docking scenarios of different diameters and materials, meeting the needs of industrial production.
Smart Images
Figure CN122134695A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent welding and three-dimensional point cloud detection technology for industrial robots, specifically to a method for detecting weld features in pipeline docking scenarios, aiming to achieve high-precision and automated detection of the geometric features of welds in the bevel area of pipelines. Background Technology
[0002] Weld inspection is a crucial step in welding quality control, directly impacting product structural safety and service life. With the development of industrial automation and intelligent manufacturing, weld inspection technology is evolving from traditional manual visual inspection and contact-based methods towards non-contact machine vision and 3D point cloud perception. Pipeline weld inspection is particularly typical and challenging in industrial production. Pipelines are widely used in petrochemical, natural gas transportation, nuclear power, and marine engineering fields, and their welding quality directly affects the safety and reliability of system operation. Pipeline welds are mostly circumferential or butt welds, typically located on cylindrical curved surfaces, exhibiting complex structures and variable spatial orientations. Traditional 2D vision or manual inspection struggles to acquire complete geometric information, resulting in low accuracy and limited efficiency. Furthermore, due to the symmetry of cylindrical structures, relying solely on pipe surface geometric features for registration during pipe butt welding and weld inspection can easily lead to local optima, hindering high-precision orientation estimation, especially in bevel regions. While markers can enhance local features, efficiently and reliably extracting markers from point clouds, constructing feature vectors, and combining them with precise registration algorithms for weld feature detection remains a challenge for current technologies. Therefore, there is an urgent need for an automated method that can combine 3D point clouds, markers, and pipeline geometric features to achieve high-precision pipeline butt weld feature detection, so as to improve detection accuracy, efficiency, and robustness. Summary of the Invention
[0003] The purpose of this invention is to provide a method for detecting weld features in pipeline docking scenarios, so as to meet the complex welding requirements of docking pipelines.
[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for detecting weld features in pipeline docking scenarios involves constructing point cloud features and using a combination of two coarse and fine registration methods to accurately dock the pipelines. The method then calculates the bevel features in the docked state, including the following steps: S1: Acquire 3D point cloud data of the source pipeline and the target pipeline. Use a 3D camera to collect the point cloud of the source pipeline carrying markers and the point cloud of the target pipeline that has been docked and carries markers. S2: Preprocess the collected point cloud data, including downsampling, filtering, and clustering to reduce the amount of point cloud data and retain the main pipeline area; S3: Based on point cloud clustering results and marker geometric feature analysis, separate and extract the three-dimensional marker point cloud features attached to the pipeline weld position; S4: Point cloud registration is performed based on marker point cloud. Two coarse registration steps are used to perform initial pose alignment of the source pipeline point cloud. Then, a fine registration algorithm is used to accurately register the source pipeline point cloud to the target pipeline point cloud, thereby completing the spatial docking of the source pipeline and the target pipeline. S5: After completing the fine registration, the pipe section is sliced based on the registered overall point cloud. At the docking position, the point cloud is sliced along the pipe axis and the geometric feature parameters of the weld area are calculated. S6: After the pipeline weld is completed, the point cloud data of the weld is captured by the area array laser 3D camera. The corresponding features are obtained based on the changes in the weld point cloud before and after welding, so as to realize the detection of the interlayer height change after the weld is welded.
[0005] Furthermore, in step S1, the acquisition of 3D point cloud data includes the following steps: S1-1: Construct point cloud pipeline models of the target pipeline and source pipelines; the target pipeline model is formed by connecting the bevels of the two source pipelines; S1-2: Geometric magnets are attached to the surface of the pipe as markers to enhance local features, and point cloud data of the pipe carrying the point cloud of the markers is obtained by scanning with a point cloud scanner.
[0006] Further, in step S2, the point cloud data undergoes preprocessing, including the following steps: S2-1: Reduce the number of point clouds by voxelization downsampling, divide the point cloud into equal-sized grids in three-dimensional space, calculate the average coordinates of all points in the grid and replace all points in the grid with it; S2-2: Outliers are filtered out by statistical filtering. For each point in the pipeline point cloud, the squared distance to its n nearest neighboring points is calculated. If a point deviates from the average value by more than a set multiple, it is identified as an outlier and removed. S2-3: The largest effective cluster is retained by Euclidean clustering, which is the main point cloud of the pipeline.
[0007] Furthermore, in steps S2-3, Euclidean clustering includes the following steps: 1) For each data point ,use Check distance data points Less than the set distance threshold All nearest neighbors; 2) For data points The same search method is used to find the nearest neighbor until no new nearest neighbor can be found. 3) This is from The region formed by the search of all points and its nearest neighbors is considered as a cluster. 4) The number of cluster points discarded is less than the set minimum number of clusters. All clusters are selected, and the remaining clusters are considered valid clusters to complete Euclidean clustering.
[0008] Furthermore, in step S3, the extraction of marker point cloud features includes the following steps: S3-1: Perform cylindrical model fitting on the overall point cloud of the docked pipeline. Separate the target model into two cylindrical models through cylindrical model fitting to obtain a composite cylindrical structure containing two cylindrical models. S3-2: For each target pipeline point cloud obtained by separation, perform the following steps: (321) Unfold the surface of the cylinder into a two-dimensional plane based on the cylindrical model; (322) Slice the unfolded planar point cloud into equally spaced slices along the circumferential direction; (323) For each slice, a region growing algorithm is used to segment it into multiple point cloud blocks; (324) Construct a topology graph based on point cloud blocks, with point cloud blocks as nodes and adjacency relationships as edges; (325) Traverse the topology graph and identify point cloud blocks located on the protruding parts of the cylinder surface with significant continuity, and determine them as the marker point cloud of the single pipeline; S3-3: Perform the same unfolding, slicing, region growing, and topology analysis process as in step S3-2 on the two source pipe point clouds obtained by the image capture, and extract the corresponding source pipe marker point clouds respectively; S3-4: Establish a one-to-one correspondence between the two target pipeline marker point clouds obtained by separating the docking pipeline and the marker point clouds of the two source pipelines, forming two sets of marker registration pairs, which are used for the subsequent two independent coarse registration processes.
[0009] Further, in step S4, the coarse registration step includes: S4-1: Perform the following processing on each target pipeline and source pipeline point cloud: (411) The cylindrical axis of the corresponding pipe is obtained by fitting the cylindrical model; (412) Calculate the maximum and minimum projection points of the point cloud along the cylindrical axis, and take the midpoint of the two as the center point of the corresponding pipe. (413) Based on the marker point cloud features extracted in step S3, the centroid of the marker point cloud is taken as the direction point; (414) Construct a bevel direction vector with the center point pointing to the bevel point cloud, so that the direction vector uniquely points to the bevel side of the pipe; S4-2: Attitude optimization based on axis rotation search. The source pipe point cloud after coarse registration is rotated at equal angles around the cylindrical axis of the target pipe. After each rotation, the mean square distance between the two point clouds is calculated using nearest neighbor search. The rotation attitude with the smallest mean square distance is selected as the optimal rotation attitude and used as the initial attitude for fine registration.
[0010] Furthermore, in step S4, the fine registration algorithm includes the following steps: Using the optimal pose obtained from the coarse registration step as the initial pose, iterative nearest-point fine registration is performed on the point cloud to obtain the final accurate aligned pose, thus realizing the registration of the pipeline docking point cloud.
[0011] Furthermore, in step S5, the geometric characteristic parameters of the weld area include: bevel face angle, half-bevel width, inner bevel height, height from the inflection point to the inner wall, blunt edge length, and root gap.
[0012] Furthermore, in step S5, the method for solving the geometric feature parameters of the weld area includes the following steps: S5-1: After fine registration, the point cloud is sliced at equal intervals along the positive Z direction, and a topology graph is constructed for each slice using the method in S3. S5-2: Find the point cloud planar relationship corresponding to the feature based on the topology map, and calculate the bevel feature parameters through the relationship between the corresponding planes.
[0013] Furthermore, the topology graph includes the following planes: upper bevel plane, inclined bevel plane, upper inclined bevel face, inclined bevel face, lower inclined bevel face, and lower bevel plane.
[0014] Compared with the prior art, the beneficial effects achieved by the present invention include: 1. High-precision weld feature extraction: By pasting marker point clouds in the pipe bevel area and combining cylindrical model fitting, point cloud slicing and topology map construction, the key geometric feature parameters of the weld can be accurately extracted, including bevel face angle, bevel width, bevel height, root gap and interlayer height variation, etc., which improves the accuracy of weld feature detection.
[0015] 2. Highly robust registration method: Based on the marker material center, feature vectors are constructed, and coarse registration and fine registration algorithms (including cylinder axis rotation optimization and iterative nearest point algorithm) are combined to effectively solve the problem that registration under cylindrical symmetric structure is prone to getting trapped in local optima, and ensure the accuracy of pipeline docking attitude estimation.
[0016] 3. Online inspection of interlayer quality after welding: By acquiring point cloud data after welding and registering it with the target bevel point cloud data before welding, the change in interlayer height of weld is accurately calculated, realizing online evaluation of interlayer quality of weld and providing a basis for subsequent process adjustments.
[0017] 4. Automation and Intelligence: The entire method is based on 3D point cloud and marker features, realizing full automation from pipeline point cloud acquisition, feature extraction, registration to weld quality inspection, without the need for manual intervention, thus improving inspection efficiency and reliability.
[0018] 5. Strong applicability: The method of this invention can be applied to pipe docking scenarios with different diameters, materials and complex shapes. It has both versatility and scalability, and meets the needs of high-precision weld inspection in industrial production. Attached Figure Description
[0019] Figure 1 This is a schematic diagram of the geometric characteristic parameters of the weld area. Detailed Implementation
[0020] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.
[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0022] This invention provides a method for detecting weld features in pipeline docking scenarios, comprising the following steps: S1: Acquire 3D point cloud data of the source and target pipes. A 3D camera is used to collect the point cloud of the source pipe segment to be docked, carrying markers, and the point cloud of the target pipe segment that has already been docked and carries markers. A handheld 3D scanner is used to perform a comprehensive, detailed scan of the docking pipe model and the single pipe model to be docked, thereby generating complete pipe point cloud information. Markers are affixed to the pipe surface near the bevel to enhance local characteristics. These markers are used to avoid the problem of registration easily getting trapped in local optima due to the perfect symmetry of the circular pipe surface, thereby improving registration accuracy.
[0023] S2: Preprocess the collected point cloud data, including downsampling, filtering, and clustering to reduce the amount of point cloud data and retain the main pipeline area; 1) First, the acquired point cloud is downsampled to reduce the amount of data and improve computational efficiency. The point cloud is divided into equal-sized grids in 3D space, and the mean coordinates of all points in the grid are calculated and used to replace all points in the grid, generating new point cloud data. The new point cloud data will largely retain the original data characteristics while significantly reducing computation time.
[0024] 2) Statistical filtering is used to filter out outliers. For each point in the pipeline point cloud, the squared distance to its n nearest neighboring points is calculated. If a point deviates from the average by more than k times, it is considered an outlier and needs to be removed. The values of n and k need to be set in advance. The matching values can be set according to actual needs. For example, n can be set to 10 and k can be set to 3. Statistical filtering can filter out most isolated points to reduce interference during data calculation.
[0025] 3) The final data preprocessing is completed using Euclidean clustering, which filters out the vast majority of invalid clusters. Euclidean clustering retains the largest valid cluster, which constitutes the main point cloud of the pipeline. A distance threshold is set. The minimum number of clusters is m.
[0026] Euclidean clustering includes the following steps: 3-1) For each data point ,use Check distance point Less than the set distance threshold All nearest neighbors; 3-2) For data points The same search method is used to find the nearest neighbor until no new nearest neighbor can be found. 3-3) Treat all points in this connected region as a single cluster; 3-4) Discard all clusters with fewer than the set minimum number of clusters m, retain the valid clusters, and complete the Euclidean clustering.
[0027] S3: Extract the marker point cloud distributed in the pipe weld groove area. Based on the point cloud clustering results and geometric feature analysis, separate and extract the three-dimensional marker point cloud attached to the pipe weld position. 1) To accurately segment the markers and improve computational efficiency, the bevel range needs to be segmented first. For the point cloud of the single bevel to be connected, the calculation is performed based on the characteristic that the normal vector of the point cloud at the bevel location changes significantly within the point cloud range at both ends of the pipeline.
[0028] For point Select its neighborhood point set , in, For neighborhood points arrive The distance.
[0029] Find the centroid of the neighborhood point set. :
[0030] Calculate the covariance matrix for the neighborhood point set:
[0031] For covariance matrix Perform eigenvalue decomposition:
[0032] Three eigenvalues were obtained and the corresponding feature vectors The eigenvector corresponding to the smallest eigenvalue. That is, the normal vector of that point:
[0033] It represents the direction with the minimum local planar variance, that is, perpendicular to the point cloud surface.
[0034] 2) After unfolding the cylindrical surface of the point cloud into a two-dimensional plane along the reference Z direction based on the cylindrical model, the point cloud surface features are obtained by using region growing. The resulting surfaces are used to establish geometric topological features for selecting feature surfaces, which is used to accurately obtain the features between the surfaces.
[0035] To obtain the marker point cloud, the bevel surface needs to be accurately located first. Therefore, the following geometric filtering relationship is established: 2-1) Surface normal vector constraint: Among the candidate surfaces obtained from region growing and segmentation, select the normal vector of the surface and the global surface. Surfaces whose axial directions are approximately the same, i.e., satisfying
[0036] in, for Axial unit vector.
[0037] 2-2) Vertical auxiliary surface constraint: The selected face should have at least one adjacent face that is approximately perpendicular to it, and the normal vector of this adjacent face should satisfy the following condition.
[0038] 2-3) Center of mass constraint The centroid of the adjacent surface is at The projection scalar along the axis is smaller than the principal candidate surface, i.e.
[0039] in These are the centroids of the primary candidate face and the adjacent face, respectively.
[0040] Among the candidate faces that simultaneously meet the above conditions, further topological consistency and scale constraints are achieved based on the geometric topological adjacency relationships between faces: the face with the largest number of point clouds and the most stable geometric structure is selected as the final bevel feature face. This topological constraint can effectively eliminate erroneously segmented faces, thereby accurately locating the region where the bevel is located.
[0041] 3) Plane fitting based on RANSAC: For a given bevel feature surface, RANSAC is used for plane fitting to obtain the plane parameters.
[0042] Marker point cloud extraction: Marker point cloud is defined as... All points whose distance from the fitting plane along the positive axis is within a specified threshold range, i.e., satisfying...
[0043] in The threshold value is used.
[0044] 4) Inverse transformation of point cloud to cylindrical coordinate system: The extracted marker point cloud is inversely transformed based on the cylindrical model parameters and mapped back to the cylindrical coordinate system to obtain the final accurate marker point cloud distribution.
[0045] S4: Point cloud registration is performed based on marker point clouds. Two coarse registration steps are used to perform initial pose alignment of the source pipeline point cloud, and then a fine registration algorithm (ICP or feature-guided registration) is used to accurately register the source pipeline point cloud to the target pipeline point cloud, thereby completing the spatial docking of the source pipeline and the target pipeline.
[0046] The core of this step is two coarse registration transformations, which use statistics to match the optimal pose.
[0047] First, extract the marker point clouds for both the source and target pipes, and calculate their feature vectors, which are denoted as coarse registration vectors. (Source pipeline) and (The two parts obtained by splitting the target pipeline on both sides). During registration, using... Two sets of coarse registration candidate attitudes are constructed based on the baseline: Candidate stance 1: Align to The initial transformation matrix is obtained.
[0048] The source point cloud after its action is recorded as follows
[0049] Candidate stance 2: Align to This yields another initial transformation matrix.
[0050] The point cloud after the action is
[0051] Next, the present invention will respectively address and Perform an axial rotation search. Constrained by the axis of the target pipe cylinder, scan the point cloud around its respective axis at different angles and calculate the corresponding mean square error (MSE).
[0052]
[0053] in, For the vector The rotation matrix is denoted by Target, where Target is the target point cloud.
[0054] This yields the minimum error under two different attitudes:
[0055]
[0056] Finally in and The coarse registration result with the smaller error is selected as the optimal initial pose. This optimal pose is then further input into an ICP or feature-guided fine registration algorithm to achieve high-precision spatial docking between the source pipe and the target pipe. The same steps are used to obtain the optimal pose for the other pipe to be registered (both single pipes require the same steps). The two steps involve aligning the two source pipes to the target pipe using cylindrical fitting to separate the two pipes, so it's a two-to-two process. Each source pipe undergoes two coarse registrations, but only one model is matched, hence the need for two coarse registrations per source pipe.
[0057] S5: After completing the fine registration, the pipe section is sliced based on the registered overall point cloud. At the docking position, the point cloud is sliced along the pipe axis and the geometric feature parameters of the weld area are calculated. like Figure 1 As shown, the weld characteristic parameters include: bevel face angles a, b, and c; half-bevel width w; inner bevel height h; height H from the inflection point to the inner wall; blunt edge length p; and root gap r.
[0058] The specific calculation steps for the geometric characteristic parameters of the weld area are as follows: First, the bevel face angle is obtained by calculating the angle between the upper plane of the bevel and the inclined plane of the bevel. Then, by calculating the angle between the inclined plane on the bevel edge and the upper plane of the bevel, the bevel face angle is obtained. Then, by calculating the angle between the lower slope of the bevel and the upper plane of the bevel, the bevel face angle is obtained. Then, calculate the intersection line between the bevel plane and the upper slope of the bevel, and measure the distance from this intersection line to the lower plane of the bevel, thereby obtaining the height from the inflection point to the inner wall. Furthermore, the length of the sloping edge is obtained by calculating the length of the intersection line between the upper and lower sloping surfaces of the sloping edge in the sloping edge plane. The inner bevel height is obtained by calculating the intersection of the lower slope and the lower slope surface and its distance to the bevel plane. Subsequently, the distance between the blunt edges of corresponding sections of the two butted pipes was measured to obtain the root gap. Finally, calculate the intersection lines of the upper plane and the inclined plane of the corresponding slice bevels, measure their distance, and divide the distance by 2 to obtain the half-bevel width. By following the steps described above, the key geometric feature parameters of the pipeline weld can be accurately obtained, providing a precise basis for the inspection and registration of pipeline butt welds.
[0059] Furthermore, to obtain the bevel edge region, this invention extracts the bevel slope plane based on the topology obtained in step S3, and transforms the registration vector using the coarse registration transformation matrix obtained in step S4, thereby obtaining the direction vector of the weld in the registered coordinate system. By filtering points in the slope plane that exceed a certain threshold along the positive direction of this direction vector, candidate regions for the upper and lower surfaces of the bevel edge can be obtained.
[0060] Because the point cloud distribution in this region is sparse and the cross-sectional features are not obvious, directly using traditional segmentation algorithms is prone to false positives and false negatives. Therefore, this invention proposes a feature extraction method of "first reducing dimensionality, then increasing dimensionality", the steps of which are as follows: Dimensional reduction projection: Projecting the sliced point cloud onto a two-dimensional plane constructed with the y-axis as the normal vector and the centroid of the sliced point cloud as the reference point, yields two-dimensional point cloud features with significant structural differences.
[0061] Sliding window search for blunt edge point clouds (feature sampling): Calculate the minimum and maximum projection points of the point cloud along the positive direction of the direction vector, set the minimum projection point as the starting position, and slide the point cloud according to the set step size; count the number of points in each sliding window, and select the segment with the most points as the edge point cloud region.
[0062] Dimensional reconstruction of the upper and lower planes of the blunt edge and the blunt edge body: Calculate the maximum and minimum values of the projection of the edge point cloud onto the Z-axis: Points whose projection value is greater than the maximum value are classified as points on the upper edge plane; Points whose projection value is less than the minimum value are classified as points on the lower edge plane. The points between the two are called the point cloud of the blunt edge body.
[0063] By following the steps above, we can obtain the three key feature surfaces corresponding to the bevel (bevel slope plane, upper edge plane, and lower edge plane). Based on these three feature surfaces, we can calculate the various bevel parameter features required in step S5. After obtaining the three feature surfaces, we can calculate the geometric feature parameters corresponding to S5.
[0064] S6: After the pipeline weld is completed, the point cloud data of the weld is captured by the area array laser 3D camera. The corresponding features, such as the change of interlayer height and the width of the weld bead, are obtained based on the change of the weld point cloud before and after welding, so as to realize the detection of the change of interlayer height after welding.
[0065] The specific steps for calculating the interlayer height of welds are as follows: The post-weld point cloud data is sliced at equal intervals along the weld direction, and surface fitting is performed on the point cloud of each slice to construct the topological relationship between the slices. Then, the height of the post-weld slice surface is compared with the pre-weld designed weld height to calculate the interlayer height variation, including parameters such as groove filling height and the spacing between the upper and lower weld layers. By analyzing the variation trend of each slice height, it is possible to determine whether the interlayer weld filling is uniform during welding and whether there are excessively high or low areas, thereby achieving online detection and evaluation of weld interlayer quality.
[0066] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for detecting weld features in a pipeline docking scenario, characterized in that, Includes the following steps: S1: Acquire 3D point cloud data of the source pipeline and the target pipeline. Use a 3D camera to collect the point cloud of the source pipeline carrying markers and the point cloud of the target pipeline that has been docked and carries markers. S2: Preprocess the collected point cloud data, including downsampling, filtering, and clustering to reduce the amount of point cloud data and retain the main pipeline area; S3: Based on point cloud clustering results and marker geometric feature analysis, separate and extract the three-dimensional marker point cloud features attached to the pipeline weld position; S4: Point cloud registration is performed based on marker point cloud. Two coarse registration steps are used to perform initial pose alignment of the source pipeline point cloud. Then, a fine registration algorithm is used to accurately register the source pipeline point cloud to the target pipeline point cloud, thereby completing the spatial docking of the source pipeline and the target pipeline. S5: After completing the fine registration, the pipe section is sliced based on the registered overall point cloud. At the docking position, the point cloud is sliced along the pipe axis and the geometric feature parameters of the weld area are calculated. S6: After the pipeline weld is completed, the point cloud data of the weld is captured by the area array laser 3D camera. The corresponding features are obtained based on the changes in the weld point cloud before and after welding, so as to realize the detection of the interlayer height change after the weld is welded.
2. The method for detecting weld features in a pipeline docking scenario according to claim 1, characterized in that, In step S1, the acquisition of 3D point cloud data includes the following steps: S1-1: Construct point cloud pipeline models of the target pipeline and source pipelines; the target pipeline model is formed by connecting the bevels of the two source pipelines; S1-2: Geometric magnets are attached to the surface of the pipe as markers to enhance local features, and point cloud data of the pipe carrying the point cloud of the markers is obtained by scanning with a point cloud scanner.
3. The method for detecting weld features in a pipeline docking scenario according to claim 1, characterized in that, In step S2, the point cloud data undergoes preprocessing, including the following steps: S2-1: Reduce the number of point clouds by voxelization downsampling, divide the point cloud into equal-sized grids in three-dimensional space, calculate the average coordinates of all points in the grid and replace all points in the grid with it; S2-2: Outliers are filtered out by statistical filtering. For each point in the pipeline point cloud, the squared distance to its n nearest neighboring points is calculated. If a point deviates from the average value by more than a set multiple, it is identified as an outlier and removed. S2-3: The largest effective cluster is retained by Euclidean clustering, which is the main point cloud of the pipeline.
4. The method for detecting weld features in a pipeline docking scenario according to claim 3, characterized in that, In steps S2-3, Euclidean clustering includes the following steps: 1) For each data point ,use Check distance data points Less than the set distance threshold All nearest neighbors; 2) For data points The same search method is used to find the nearest neighbor until no new nearest neighbor can be found. 3) This is from The region formed by the search of all points and its nearest neighbors is considered as a cluster. 4) The number of cluster points discarded is less than the set minimum number of clusters. All clusters are selected, and the remaining clusters are considered valid clusters to complete Euclidean clustering.
5. The method for detecting weld features in a pipeline docking scenario according to claim 1, characterized in that, Step S3, the extraction of marker point cloud features, includes the following steps: S3-1: Perform cylindrical model fitting on the overall point cloud of the docked pipeline. Separate the target model into two cylindrical models through cylindrical model fitting to obtain a composite cylindrical structure containing two cylindrical models. S3-2: For each target pipeline point cloud obtained by separation, perform the following steps: (321) Unfold the surface of the cylinder into a two-dimensional plane based on the cylindrical model; (322) Slice the unfolded planar point cloud into equally spaced slices along the circumferential direction; (323) For each slice, a region growing algorithm is used to segment it into multiple point cloud blocks; (324) Construct a topology graph based on point cloud blocks, with point cloud blocks as nodes and adjacency relationships as edges; (325) Traverse the topology graph and identify point cloud blocks located on the protruding parts of the cylinder surface with significant continuity, and determine them as the marker point cloud of the single pipeline; S3-3: Perform the same unfolding, slicing, region growing, and topology analysis process as in step S3-2 on the two source pipe point clouds obtained by the image capture, and extract the corresponding source pipe marker point clouds respectively; S3-4: Establish a one-to-one correspondence between the two target pipeline marker point clouds obtained by separating the docking pipeline and the marker point clouds of the two source pipelines, forming two sets of marker registration pairs, which are used for the subsequent two independent coarse registration processes.
6. The method for detecting weld features in a pipeline docking scenario according to claim 1, characterized in that, In step S4, the coarse registration step includes: S4-1: Perform the following processing on each target pipeline and source pipeline point cloud: (411) The cylindrical axis of the corresponding pipe is obtained by fitting the cylindrical model; (412) Calculate the maximum and minimum projection points of the point cloud along the cylindrical axis, and take the midpoint of the two as the center point of the corresponding pipe. (413) Based on the marker point cloud features extracted in step S3, the centroid of the marker point cloud is taken as the direction point; (414) Construct a bevel direction vector with the center point pointing to the bevel point cloud, so that the direction vector uniquely points to the bevel side of the pipe; S4-2: Attitude optimization based on axis rotation search. The source pipe point cloud after coarse registration is rotated at equal angles around the cylindrical axis of the target pipe. After each rotation, the mean square distance between the two point clouds is calculated using nearest neighbor search. The rotation attitude with the smallest mean square distance is selected as the optimal rotation attitude and used as the initial attitude for fine registration.
7. The method for detecting weld features in a pipeline docking scenario according to claim 1, characterized in that, In step S4, the fine registration algorithm includes the following steps: Using the optimal pose obtained from the coarse registration step as the initial pose, iterative nearest-point fine registration is performed on the point cloud to obtain the final accurate aligned pose, thus realizing the registration of the pipeline docking point cloud.
8. The method for detecting weld features in a pipeline docking scenario according to claim 1, characterized in that, In step S5, the geometric characteristic parameters of the weld area include: bevel face angle, half-bevel width, inner bevel height, height from the inflection point to the inner wall, blunt edge length, and root gap.
9. The method for detecting weld features in a pipeline docking scenario according to claim 1, characterized in that, In step S5, the method for solving the geometric feature parameters of the weld area includes the following steps: S5-1: After fine registration, the point cloud is sliced at equal intervals along the positive Z direction, and a topology graph is constructed for each slice using the method in S3. S5-2: Find the point cloud planar relationship corresponding to the feature based on the topology map, and calculate the bevel feature parameters through the relationship between the corresponding planes.
10. The method for detecting weld features in a pipeline docking scenario according to claim 1, characterized in that, The topology graph includes the following planes: upper bevel plane, bevel inclined plane, upper bevel face, bevel face, lower bevel face, and lower bevel plane.