A method for generating 3D models of oil and gas pipelines based on point cloud reconstruction
By using multi-site point cloud registration and an improved PF-Net network, combined with the pipeline cross-section status and branch opening and closing relationships, the problems of pipe wall misalignment and structural fracture in the generation of 3D models of oil and gas pipelines were solved, and stable, complete and continuous reconstruction of 3D models of oil and gas pipelines was achieved.
Patent Information
- Application Number
- CN202611523321.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-29
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for generating 3D models of oil and gas pipelines are prone to positional offsets and normal differences when scanning multiple sites, leading to pipe wall misalignment, cross-sectional center drift, or local structural fractures. In particular, it is difficult to accurately recover missing point clouds in areas where pipes are bent, reduce diameter, or branches intersect, affecting the structural integrity of the model.
By employing multi-site point cloud registration and continuous connection, pipeline cross-section chain construction, improved PF-Net network point cloud completion, pipe wall mesh reconstruction, and pipe segment topology connection technology, a three-dimensional model with continuous pipe wall structure and pipe segment connection relationship is generated through point cloud preprocessing, continuous point cloud construction, marking of gap and branch intersection areas, generation of missing pipe wall points, and pipe segment segmentation.
It improves the stability of the conversion of point cloud data of oil and gas pipelines from discrete acquisition data to continuous three-dimensional structure, reduces pipe wall misalignment caused by changes in scanning perspective and differences in inter-station location, enhances the structural continuity and geometric integrity of the model, and ensures clear branch relationships and continuous transition of pipe segment boundaries.
Smart Images

Figure CN123118922A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of 3D modeling technology, and in particular to a method for generating 3D models of oil and gas pipelines based on point cloud reconstruction. Background Technology
[0002] As oil and gas pipeline networks continue to expand, pipeline routes, diameter variations, and branch connection methods become increasingly complex. Accurately acquiring the spatial structure of pipelines and generating 3D models has become a crucial foundation for digital pipeline management. Existing methods for generating 3D models of oil and gas pipelines typically utilize laser scanning equipment or 3D measurement equipment to collect point clouds from multiple scanning stations. These point clouds are then denoised, coordinate-processed, registered between stations, and stitched together. The pipeline surface structure is then restored through surface fitting, mesh reconstruction, or geometric parameter extraction. For straight pipes, bends, pipes with varying diameters, and branch connection areas, the center trajectory, cross-sectional dimensions, and local extension directions can be extracted based on the spatial distribution of the point cloud to construct pipe wall meshes and segment connection relationships, thus forming a 3D model of the oil and gas pipeline.
[0003] Existing point cloud-based methods for generating 3D models of oil and gas pipelines largely rely on registration using overlapping points from adjacent scanning stations, and employ neighborhood interpolation, surface fitting, or general point cloud completion methods to repair missing areas. Multi-station scanning is affected by viewing angle, occlusion, and inter-station positions, leading to positional shifts and normal differences in overlapping pipe segments, resulting in pipe wall misalignment, cross-sectional center drift, or local structural fractures. Point cloud gaps at bends, reducers, and branch intersections are accompanied by changes in cross-sectional direction, radial scale, and opening status. Existing completion methods struggle to utilize the structural changes on both sides of the gap to constrain the generation of missing points, easily resulting in discontinuous center trajectories, cross-sectional scales, or branch openings. When directly performing surface meshing after point cloud completion, the correspondence and adjacency relationships between pipe segment ports lack further verification, and mesh misconnections, boundary non-closures, or transition discontinuities easily occur in boundary areas, affecting the structural integrity of the 3D oil and gas pipeline model.
[0004] Therefore, how to provide a method for generating 3D models of oil and gas pipelines based on point cloud reconstruction is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] One objective of this invention is to propose a method for generating 3D models of oil and gas pipelines based on point cloud reconstruction. This invention utilizes multi-site point cloud registration and continuation, pipeline cross-section chain construction, improved PF-Net network point cloud completion, pipe wall mesh reconstruction, and pipe segment topology connection technology. It details the implementation process of multi-site point cloud preprocessing, continuous point cloud construction, point cloud gap and branch intersection area marking, missing pipe wall point generation, pipe segment segmentation, and boundary pipe wall reconstruction for oil and gas pipelines. This generates a 3D model of an oil and gas pipeline with continuous pipe wall structure and pipe segment connection relationships. Compared to traditional point cloud interpolation, surface fitting, and direct mesh reconstruction methods, this invention combines pipeline cross-section state, structural changes on both sides of gaps, and branch opening / closing relationships to constrain the point cloud completion process. It has advantages such as stable multi-site point cloud continuation, accurate restoration of missing pipe wall structures, complete reconstruction of branch intersection areas, and clear pipe segment connection relationships.
[0006] A method for generating a 3D model of an oil and gas pipeline based on point cloud reconstruction according to an embodiment of the present invention includes: Collect point clouds from multiple sites along oil and gas pipelines and perform preprocessing to generate a standard point cloud set for oil and gas pipelines. Based on the standard point cloud of oil and gas pipelines, the overlapping pipe segments of adjacent scanning stations are paired, the pose between stations is corrected and the connection is continued along the pipeline extension direction to generate a continuous point cloud of oil and gas pipelines. Divide the pipeline cross-section along the continuous point cloud of the oil and gas pipeline, connect adjacent cross-sections in sequence to form a cross-section chain, mark the point cloud gaps and branch intersection areas, and generate oil and gas pipeline reconstruction samples. The oil and gas pipeline reconstruction sample is input into the improved PF-Net network, which includes a pipeline state reinjection unit, a gap recursive anchoring unit, a pipe wall hierarchical generation unit, and a cross-section opening and closing migration unit to generate a complete point cloud of the oil and gas pipeline. Based on the extraction of the center trajectory from the complete point cloud of the oil and gas pipeline, the complete point cloud of the oil and gas pipeline is projected onto the normal section of the center trajectory and rearranged axially and circumferentially to construct the pipe wall mesh and divide the pipe segments, thereby generating the reconstructed structure of the oil and gas pipeline. Based on the reconstructed structure of the oil and gas pipeline, the pipe segment connection ports are extracted, the port boundaries are matched and the pipe segment adjacency relationship is established to generate a three-dimensional topology model of the oil and gas pipeline. Based on the 3D topology model of the oil and gas pipeline, the boundary wall of adjacent pipe segments is cut, the cut boundary is reorganized and a transition mesh is constructed to generate the 3D model of the oil and gas pipeline.
[0007] Optionally, the point cloud of the oil and gas pipeline multi-site includes point clouds of straight pipe sections, point clouds of bends, point clouds of pipe sections with varying diameters, point clouds of branch intersections, and point clouds of pipe fitting connection areas.
[0008] Optionally, the generation of the oil and gas pipeline standard point set includes: Divide the oil and gas pipeline multi-site point cloud according to the scanning station, calculate the spatial distance between each point and its neighboring points and remove outliers to generate a denoised point cloud of the oil and gas pipeline. Unify the coordinate units and coordinate axis arrangement rules of the denoised point cloud of oil and gas pipelines, without changing the original relative pose between each scanning station, and generate a standard point cloud of oil and gas pipelines. For each point in the standard point cloud of oil and gas pipelines, neighborhood points are selected, and the local covariance matrix is calculated based on the coordinates of the neighborhood points. The eigenvector corresponding to the minimum eigenvalue is determined as the point normal. The three-dimensional coordinates, point normals, and scanning station identifiers of the standard point cloud of oil and gas pipelines are encapsulated to generate a standard point cloud set of oil and gas pipelines.
[0009] Optionally, generating a continuous point cloud of oil and gas pipelines includes: The standard point cloud of oil and gas pipelines is divided according to the scanning station identification. The point clouds of adjacent scanning stations are paired. Corresponding points are selected according to the spatial distance between points and the minimum angle between the point normals. The pipe wall area where the corresponding points are continuously distributed is determined as the overlapping pipe section. Based on the three-dimensional coordinates and point normals of corresponding points within the overlapping pipe sections, the rotation and translation between adjacent scanning stations are calculated. The point cloud pose of the corresponding scanning station is corrected according to the rotation and translation, and the normal of the corresponding point is corrected synchronously according to the rotation. According to the pairing order of adjacent scanning stations, the overlapping pipe segments after pose correction are used as the continuation area, and the non-overlapping pipe segments are spliced sequentially along the extension direction of the oil and gas pipeline to generate the oil and gas pipeline continuation point cloud. Merge the corresponding points and corrected point normals of overlapping pipe segments in the oil and gas pipeline connection point cloud, and retain the pipe wall points and corrected point normals of non-overlapping pipe segments to generate a continuous point cloud of oil and gas pipelines.
[0010] Optionally, the process of forming the cross-section chain includes: The continuous point cloud along the oil and gas pipeline is divided into local pipe wall point groups. The local pipeline extension direction is determined according to the point normal of each local pipe wall point group, and the pipeline section is set with the local pipeline extension direction as the section normal. Project the pipe wall points in each local pipe wall point group onto the corresponding pipe section, determine the section center according to the spatial distribution of the projection points, and encapsulate the pipe wall points, projection points, corresponding point normals, section center and section normals into a pipe section; Using each pipe section as the current section, the center of the adjacent pipe sections is projected onto the local pipe extension direction of the current section. Based on the projection position, the preceding and subsequent sections are determined to generate a pipe section sequence. Connect the preceding and succeeding sections according to the pipe section sequence. For pipe sections with multiple preceding or succeeding sections, retain each connection branch and record the connection sequence of the pipe sections to form a section chain.
[0011] Optionally, generating the oil and gas pipeline reconstruction sample includes: Calculate the center-to-center distance between adjacent pipe sections along the cross-section chain, compare each center-to-center distance with the median of the center-to-center distances between adjacent pipe sections in the cross-section chain, and mark the point cloud gaps between adjacent pipe sections whose center-to-center distance is greater than twice the median. Count the number of adjacent pipe sections connected to each pipe section along the cross-section chain, and mark the pipe sections that connect three or more adjacent pipe sections and their adjacent pipe wall areas as branch intersection areas; According to the connection sequence of the cross-section chain, the point cloud gap and the branch intersection area are respectively associated with the corresponding pipe cross-section, and the two boundary cross-sections of the point cloud gap and the connection cross-section of the branch intersection area are marked. The cross-section chain, point cloud gap, branch intersection area, two side boundary sections and connecting sections are encapsulated to generate oil and gas pipeline reconstruction samples.
[0012] Optionally, generating a complete point cloud of the oil and gas pipeline includes: An improved PF-Net network is constructed by adding a pipeline state reinjection unit, a gap recursive anchoring unit, and a pipe wall classification generation unit to the PF-Net network, and embedding a cross-section opening and closing migration unit in the pipe wall classification generation unit. The improved PF-Net network is trained by using the pipe wall reconstruction error, notch anchoring closure error, cross-section opening and closing state transition error and pipe wall point holding error as the joint loss. The trainable parameters of the improved PF-Net network are updated in reverse. When the absolute value of the difference between the joint loss of two adjacent rounds of validation sets is less than 0.0005 for 6 consecutive rounds, the improved PF-Net network is considered to have completed convergence training. The reconstructed samples of oil and gas pipelines are input into the pipeline status reinjection unit. The spatial distribution of the cross-section center, cross-section normal and projection points of each pipeline cross-section is aggregated along the cross-section chain. The center displacement, cross-section normal rotation angle and radial scale change of adjacent pipeline cross-sections are calculated. The calculation results are backfilled to the corresponding pipeline cross-sections, and the corresponding pipe wall points are retained. The point cloud gaps and branch intersection areas are associated, and the pipeline status feature sequence is output. Input the pipeline state feature sequence into the gap recursive anchoring unit, push candidate anchor points outward from the two boundary sections of the point cloud gap into the gap, eliminate unclosed candidate branches according to the center displacement, cross section normal rotation angle and radial scale change, recursively push out the remaining candidate anchor points and connect them to form a gap anchoring chain, backfill the gap anchoring chain into the pipeline state feature sequence, and output the gap anchoring feature sequence. The gap anchoring feature sequence is input into the cross-section opening and closing migration unit in the pipe wall hierarchical generation unit. The cross-section state of the intersection area of the cross-section chain and the gap anchoring chain migration branch is used to mark the corresponding pipe cross-sections as the opening entry state, the opening holding state, and the opening exit state in sequence. The remaining pipe cross-sections are marked as the closed state. The circumferential generation interval is divided according to the relative position of the cross-section center and the cross-section normal between the connecting cross-sections. The cross-section state and the circumferential generation interval are backfilled into the gap anchoring feature sequence, and the cross-section migration feature sequence is output. The cross-section migration feature sequence is input into the pipe wall hierarchical generation unit. The cross-section parent node is generated by the anchor point in the gap anchoring chain. The pipe wall sub-points are expanded step by step from the cross-section parent node according to the cross-section state and the circumferential generation interval. The pipe wall sub-points are merged with the pipe wall points retained in the cross-section migration feature sequence to generate a complete point cloud of the oil and gas pipeline.
[0013] Optionally, the generated oil and gas pipeline structure reconstruction includes: Divide the complete point cloud of the oil and gas pipeline into local pipe wall point groups, calculate the center position of each local pipe wall point group, and connect adjacent center positions according to spatial adjacency to generate the center trajectory; Set up trajectory sampling points along the central trajectory, and construct the corresponding normal section using the trajectory tangent at each trajectory sampling point as the normal. Project the complete point cloud of the oil and gas pipeline onto the corresponding normal section, and determine the axial sequence according to the arrangement order of the trajectory sampling points. Using the trajectory sampling points of each normal section as the center of the section, the projection points in the normal section are arranged according to the circumferential position of the projection points around the center of the section. The axial sequence and the circumferential sequence are combined to form an axial-circumferential pipe wall lattice, and adjacent pipe wall points are connected to form a pipe wall grid. The branch positions of the central trajectory and the changes in the grid connection between adjacent normal sections are identified along the pipe wall grid. The pipe wall grid and the central trajectory are simultaneously segmented at the corresponding positions to form pipe segments and corresponding central trajectory segments. The pipe wall grid of each pipe segment and the corresponding central trajectory segment are encapsulated to generate the reconstructed structure of the oil and gas pipeline.
[0014] Optionally, generating the three-dimensional topology model of the oil and gas pipeline includes: Based on the pipe wall mesh of each pipe segment in the reconstructed oil and gas pipeline structure, continuously distributed boundary nodes at the boundary of each pipe segment are extracted to form the port boundary. The port boundary center is calculated, and the port normal is determined according to the end extension direction of the corresponding center trajectory segment. The port boundary, port boundary center and port normal are combined to form the pipe segment connection port. Calculate the center distance between the port boundaries, the included angle of the port normal, and the difference in the perimeter of the port boundaries between the connection ports of different pipe segments. Based on the three calculation results, filter and pair the corresponding port boundaries to generate port pairing records. Associating corresponding pipe segments with port pairing records, establishing adjacency connections for corresponding pipe segments whose port boundaries have been paired, and retaining the adjacency branches corresponding to the connecting ports of each pipe segment for cases where the same pipe segment corresponds to multiple port pairing records, thus forming pipe segment adjacency relationships; By associating the pipe wall mesh, pipe connection ports, port pairing records, and pipe adjacency relationships of each pipe segment, a three-dimensional topology model of the oil and gas pipeline is generated.
[0015] Optionally, generating the three-dimensional model of the oil and gas pipeline includes: Adjacent pipe segments are determined based on the adjacency relationship of pipe segments in the three-dimensional topology model of oil and gas pipelines, the pipe wall mesh corresponding to the adjacent pipe segments is extracted, and the boundary pipe wall between the pipe segment connection ports is located. Cut the junction pipe wall along the port boundary of the pipe segment connection port, extract the boundary nodes of the cut boundaries on both sides, and sort them according to the circumferential position of the boundary nodes around the center of the corresponding port boundary. Pair the corresponding boundary nodes in the two sides of the clipping boundary according to the circumferential order, and insert transition nodes for clipping boundaries with inconsistent numbers of boundary nodes to form a clipping boundary sequence with consistent numbers of nodes on both sides. Connect corresponding nodes and adjacent nodes according to the node pairing relationship of the cut boundary sequence, construct a transition mesh that spans both cut boundaries, and merge the transition mesh with the pipe wall mesh of the adjacent pipe segment to generate a three-dimensional model of the oil and gas pipeline.
[0016] The beneficial effects of this invention are: This invention proposes a method for generating 3D models of oil and gas pipelines based on point cloud reconstruction. By constructing a multi-site point cloud registration and continuous splicing mechanism, organizing pipeline cross-sections, marking point cloud gaps and branch intersection areas, and implementing a 3D structure reconstruction process, this method improves the stability of converting complex oil and gas pipeline point clouds from discretely acquired data to continuous 3D structures. Compared to traditional methods that rely solely on local overlapping points for registration and direct stitching, this invention combines the spatial distance between points, the minimum angle between point normals, the cross-section center, and the cross-section normal to continuously organize the pipe wall point clouds from multiple sites. This reduces pipe wall misalignment and cross-section drift caused by changes in scanning perspective, occlusion, and differences in inter-site positions, thereby improving the spatial consistency of continuous point clouds in oil and gas pipelines.
[0017] This invention provides structured completion of missing regions in oil and gas pipeline point clouds by improving the PF-Net network. During pipeline state refilling, gap recursive anchoring, cross-section opening / closing migration, and pipe wall hierarchical generation, it introduces the center displacement of adjacent pipe cross-sections, cross-section normal rotation angle, radial scale changes, and the cross-section state of branch intersection areas to constrain the generation process of missing pipe wall points. Compared to neighborhood interpolation, surface fitting, and general point cloud completion methods, this invention can recover missing regions by combining the structural changes of pipelines on both sides of the gap and the opening / closing relationship of branches. It reduces abrupt changes in cross-section scale, center trajectory offset, and branch opening distortion in bends, reducing the occurrence of such abrupt changes in cross-section scale, center trajectory offset, and branch opening distortion in branch intersection areas, thereby improving the structural continuity and geometric integrity of the complete point cloud of oil and gas pipelines.
[0018] This invention further constructs a central trajectory, axial-circumferential pipe wall lattice, and pipe wall mesh based on the complete point cloud of an oil and gas pipeline. A three-dimensional topological model of the oil and gas pipeline is established through pipe segment connection ports, port boundary pairing, and pipe segment adjacency relationships. The boundary pipe walls of adjacent pipe segments are trimmed, boundaries are reorganized, and transition meshes are constructed. This process reduces problems such as pipe segment connection misalignment, boundary non-closure, and local transition discontinuities caused by direct meshing. The final generated three-dimensional oil and gas pipeline model simultaneously maintains the pipe wall geometry and pipe segment topological connection relationships, possessing advantages such as stable point cloud continuity, complete restoration of missing structures, clear expression of branch relationships, and continuous transitions at pipe segment boundaries. Attached Figure Description
[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for generating a 3D model of an oil and gas pipeline based on point cloud reconstruction proposed in this invention. Figure 2 This is a schematic diagram of the improved PF-Net network structure for a point cloud-based method for generating 3D models of oil and gas pipelines, as proposed in this invention. Figure 3 This is a flowchart illustrating the generation of a three-dimensional topological model of an oil and gas pipeline based on point cloud reconstruction, as proposed in this invention. Detailed Implementation
[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0021] refer to Figure 1 , Figure 2 and Figure 3 A method for generating 3D models of oil and gas pipelines based on point cloud reconstruction, comprising: Collect point clouds from multiple sites along oil and gas pipelines and perform preprocessing to generate a standard point cloud set for oil and gas pipelines. Based on the standard point cloud of oil and gas pipelines, the overlapping pipe segments of adjacent scanning stations are paired, the pose between stations is corrected and the connection is continued along the pipeline extension direction to generate a continuous point cloud of oil and gas pipelines. Divide the pipeline cross-section along the continuous point cloud of the oil and gas pipeline, connect adjacent cross-sections in sequence to form a cross-section chain, mark the point cloud gaps and branch intersection areas, and generate oil and gas pipeline reconstruction samples. The oil and gas pipeline reconstruction sample is input into the improved PF-Net network, which includes a pipeline state reinjection unit, a gap recursive anchoring unit, a pipe wall hierarchical generation unit, and a cross-section opening and closing migration unit to generate a complete point cloud of the oil and gas pipeline. Based on the extraction of the center trajectory from the complete point cloud of the oil and gas pipeline, the complete point cloud of the oil and gas pipeline is projected onto the normal section of the center trajectory and rearranged axially and circumferentially to construct the pipe wall mesh and divide the pipe segments, thereby generating the reconstructed structure of the oil and gas pipeline. Based on the reconstructed structure of the oil and gas pipeline, the pipe segment connection ports are extracted, the port boundaries are matched and the pipe segment adjacency relationship is established to generate a three-dimensional topology model of the oil and gas pipeline. Based on the 3D topology model of the oil and gas pipeline, the boundary wall of adjacent pipe segments is cut, the cut boundary is reorganized and a transition mesh is constructed to generate the 3D model of the oil and gas pipeline.
[0022] In this embodiment, the point cloud of the oil and gas pipeline multi-site specifically includes the point cloud of straight pipe sections, the point cloud of bends, the point cloud of pipe sections with varying diameters, the point cloud of branch intersections, and the point cloud of pipe fitting connection areas.
[0023] In this embodiment, the generation of the standard point set for oil and gas pipelines includes: The oil and gas pipeline multi-site point cloud is divided according to the scanning stations. The spatial distance between each point and its neighboring points is calculated and outliers are removed to generate a denoised point cloud of the oil and gas pipeline. Generating a denoised point cloud for oil and gas pipelines, specifically: The point clouds of multiple stations of oil and gas pipelines are grouped according to the scanning station identifier. Pipe wall points with the same scanning station identifier are grouped into the same scanning station cloud, while keeping the original three-dimensional coordinates unchanged. For each tube wall point in the cloud of each scanning station, calculate its three-dimensional Euclidean distance with the other tube wall points in the same scanning station, arrange them in ascending order of distance, and select the 20 nearest tube wall points as neighborhood points. The three-dimensional Euclidean distances from the corresponding tube wall point to 20 neighboring points are summed and divided by 20 to obtain the average neighboring distance. The average neighboring distances of each tube wall point within the same scanning station are sorted and the median of the average neighboring distances is taken. Calculate the absolute value of the difference between each average neighborhood distance and the median of the average neighborhood distance. Sort the obtained absolute values and take the median as the neighborhood distance dispersion. Determine the pipe wall points whose average neighborhood distance is greater than the sum of the median of the average neighborhood distance and 3 times the neighborhood distance dispersion as outliers. Outliers in each scanning station cloud are removed, and the remaining pipe wall points, scanning station identifiers, coordinate units and coordinate axis fields are retained and merged to generate a denoised point cloud for oil and gas pipelines. The coordinate units and coordinate axis arrangement rules of the denoised point cloud for oil and gas pipelines are standardized without changing the original relative poses between scanning stations, thus generating a standardized point cloud for oil and gas pipelines. Generate a standard point cloud for oil and gas pipelines, specifically as follows: The coordinate units and coordinate axis fields of the denoised point cloud of the oil and gas pipeline are extracted according to the scanning station identification. The conversion factor is set according to the coordinate units, and the three-dimensional coordinates of each pipe wall point are uniformly converted to millimeters. Following a fixed arrangement order of the X, Y, and Z axes, the coordinate components representing the same coordinate axis in each scanning station cloud are mapped to the corresponding coordinate fields; Only coordinate unit conversion and coordinate component mapping are performed. No inter-station rotation or translation transformation is applied to each scanning station cloud. The processed scanning station clouds are merged according to the scanning station identifier to generate the standard point cloud of oil and gas pipeline. For each point in the standard point cloud of oil and gas pipelines, neighborhood points are selected. Based on the coordinates of these neighborhood points, the local covariance matrix is calculated. The eigenvector corresponding to the smallest eigenvalue is then determined as the point normal. Where: The local covariance matrix is calculated as follows: For each pipe wall point in the standard point cloud of oil and gas pipelines, calculate its three-dimensional Euclidean distance with other pipe wall points in the same scanning station, arrange them in ascending order of distance, and select the 20 nearest pipe wall points as neighborhood points. Accumulate the X, Y, and Z coordinates of 20 neighboring points and divide by 20 to obtain the mean X, Y, and Z coordinates of the neighborhood center. Calculate the difference between the X, Y, and Z coordinates of each neighboring point and the mean of the corresponding coordinates. Sum the squares of the coordinate differences in the three directions and divide by 20 to obtain the coordinate variances in the X, Y, and Z directions. Multiply the differences in X and Y coordinates, X and Z coordinates, and Y and Z coordinates of each neighboring point respectively, accumulate the results of each multiplication, and divide by 20 to obtain the coordinate covariances in the XY, XZ, and YZ directions. The three coordinate variances are used as the main diagonal terms, and the three coordinate covariances are filled into the corresponding non-main diagonal positions to form a 3×3 local covariance matrix. Eigenvalue decomposition is performed on the local covariance matrix to determine the smallest eigenvalue among the three eigenvalues. The eigenvector corresponding to the smallest eigenvalue is normalized and the normalized eigenvector is determined as the point normal of the corresponding pipe wall point. The three-dimensional coordinates, point normals, and scanning station identifiers of the standard point cloud of oil and gas pipelines are encapsulated to generate a standard point cloud set of oil and gas pipelines.
[0024] In this embodiment, generating a continuous point cloud of oil and gas pipelines includes: The standard point cloud of oil and gas pipelines is divided according to the scanning station identification. The point clouds of adjacent scanning stations are paired, and corresponding points are selected based on the spatial distance between points and the minimum angle between the point normals. The pipe wall areas where corresponding points are continuously distributed are identified as overlapping pipe segments, where: The overlapping pipe sections are identified as follows: The standard point clusters of oil and gas pipelines are divided according to the arrangement order of the scanning station identifiers, and two scanning station clouds with adjacent arrangement positions are formed into scanning station cloud pairs. For each scanning station cloud pair, the three-dimensional Euclidean distance between the tube wall point in one scanning station and each tube wall point in another scanning station is calculated based on the tube wall point in one scanning station. The tube wall point with the smallest distance is determined, and the tube wall points that are the smallest distance points to each other are retained as candidate corresponding points. Calculate the three-dimensional Euclidean distance between each candidate corresponding point, sort the obtained distances and take the median, calculate the absolute value of the difference between each distance and the median and take the median as the candidate distance discrepancy, and determine the upper limit of distance screening by the sum of the median distance between the candidate corresponding points and 3 times the candidate distance discrepancy. Calculate the absolute value of the dot product of the normals of the two unit points of the candidate corresponding point, and perform an inverse cosine operation to obtain the minimum angle between the point normals. The candidate corresponding points whose three-dimensional Euclidean distance is not greater than the upper limit of the distance screening and whose minimum angle between the point normals is not greater than 15° are determined as the corresponding points. For each of the two scanning stations, the distance between each corresponding point and the nearest corresponding point is calculated. The median of the obtained distances is taken as the adjacency distance of the corresponding scanning station. When the distance between the two sets of corresponding points in the two scanning stations is not greater than twice the corresponding adjacency distance, an adjacency connection is established. The areas that are interconnected and contain no less than 30 sets of corresponding points are defined as overlapping pipe segments, and the three-dimensional coordinates and point normals of each set of corresponding points are retained. Based on the 3D coordinates and point normals of corresponding points within the overlapping pipe sections, the rotation and translation between adjacent scanning stations are calculated. The point cloud pose of the corresponding scanning stations is corrected according to the rotation and translation, and the point normals are synchronously corrected according to the rotation. Correcting the corresponding point cloud pose and point normals, specifically: According to the order of the scanning site cloud pairs, the first scanning site cloud is determined as the reference point cloud, and the next adjacent scanning site cloud is determined as the point cloud to be corrected. The three-dimensional coordinates and point normals of each group of corresponding points in the overlapping pipe section are extracted. Based on the current reference point cloud and the point cloud to be corrected, recalculate the three-dimensional Euclidean distance and the minimum angle of the point normal for each group of corresponding points. Divide the three-dimensional Euclidean distance by the upper limit of the distance filtering of the corresponding scanned point cloud pair, divide the minimum angle of the point normal by 15°, add the two ratios to 1 and take the reciprocal as the initial weight, and then divide by the sum of all the initial weights to obtain the corresponding point weight. Multiply the 3D coordinates of each corresponding point in the reference point cloud and the point cloud to be calibrated by the corresponding point weights and sum them up to obtain the coordinate center of the reference point and the coordinate center of the point to be calibrated. Subtract the corresponding coordinate center from the 3D coordinates of each corresponding point to obtain two sets of centered coordinates. The corresponding centered coordinates of the point cloud to be corrected and the reference point cloud are externally multiplied one by one, and then multiplied by the corresponding point weights and accumulated to form a 3×3 cross-covariance matrix. The cross-covariance matrix is then subjected to singular value decomposition to obtain the left singular vector matrix and the right singular vector matrix. Multiply the right singular vector matrix by the transpose of the left singular vector matrix. If the determinant of the resulting matrix is less than 0, multiply the last column of the right singular vector matrix by -1 and then multiply again to obtain the rotation matrix and determine it as the rotation amount. The coordinate center of the point to be corrected is transformed by using a rotation matrix. The difference between the coordinate center of the reference point and the coordinate center of the transformed point to be corrected is used to obtain the three-dimensional translation vector and determine it as the translation amount. Multiply the three-dimensional coordinates of each pipe wall point in the point cloud to be corrected by the rotation matrix, then add the three-dimensional translation vector, and multiply the point normal of each pipe wall point by the rotation matrix and normalize it. The next adjacent scan point cloud after the calibration is completed is used as the reference point cloud for the next scan point cloud pair, and all scan points are calibrated in the pairing order. Following the pairing order of adjacent scanning stations, and using the overlapping pipe segments after pose correction as the splicing region, the non-overlapping pipe segments are sequentially spliced along the extension direction of the oil and gas pipeline to generate a splicing point cloud of the oil and gas pipeline, wherein: Generate point clouds for oil and gas pipeline connections, specifically as follows: Arrange the cloud of each scanning station that has completed pose correction according to the pairing order of the scanning stations, and merge the pipe wall points in each group of adjacent scanning stations that have not been included in the overlapping pipe segment according to the spatial connectivity relationship to form a non-overlapping pipe segment. Calculate the mean coordinates of the pipe wall points in the overlapping pipe segment and the corresponding non-overlapping pipe segment respectively to obtain the center of the overlapping pipe segment and the center of the non-overlapping pipe segment. Normalize the vector pointing from the center of the overlapping pipe segment to the center of the non-overlapping pipe segment to obtain the extension direction of the corresponding oil and gas pipeline. The first scanning station cloud is used as the initial connection point cloud. The corrected three-dimensional coordinates and point normals of the non-overlapping pipe section of the next adjacent scanning station are retained. The connection is made according to the projection position of the non-overlapping pipe section relative to the center of the overlapping pipe section along the extension direction of the oil and gas pipeline. According to the pairing order of scanning stations, the overlapping pipe section between the already connected scanning station and the next adjacent scanning station is taken as the connection area, and the corresponding non-overlapping pipe sections are connected in sequence to generate the oil and gas pipeline connection point cloud. The corresponding points and corrected point normals of overlapping pipe segments in the oil and gas pipeline connection point cloud are merged, while the pipe wall points and corrected point normals of non-overlapping pipe segments are retained to generate a continuous point cloud of oil and gas pipelines, wherein: Generate a continuous point cloud of oil and gas pipelines, specifically as follows: According to the correspondence within the overlapping pipe sections, pipe wall points that can be continuously associated through the correspondence are grouped into the same corresponding point group; The corrected X, Y, and Z coordinates of all pipe wall points in each corresponding point group are summed and divided by the number of corresponding points to obtain the merged pipe wall points. Using the normal of the first point in the corresponding point group as the direction reference, calculate its dot product with the normals of the other points. Multiply the three components of the normal of the points whose dot product is less than 0 by -1. Then, sum the corresponding components of the normal of each point and normalize them to obtain the merged point normal. Replace the pipe wall points in the corresponding point group with the merged pipe wall points and merged point normals, retain the corrected pipe wall points and point normals of each non-overlapping pipe segment, and merge them according to the scanning station connection order to generate a continuous point cloud of oil and gas pipeline.
[0025] In this embodiment, forming the cross-sectional chain includes: The continuous point cloud along the oil and gas pipeline is divided into local pipe wall point groups. The local pipeline extension direction is determined based on the point normal of each local pipe wall point group, and the pipeline section is set with the local pipeline extension direction as the section normal. The pipe cross-section is set as follows: For each pipe wall point in the continuous point cloud of oil and gas pipelines, calculate its three-dimensional Euclidean distance with the other pipe wall points, arrange them in ascending order of distance, select the 100 nearest pipe wall points as the center of the corresponding pipe wall point as the initial neighborhood, and continue to add pipe wall points in order of distance. Calculate the absolute value of the dot product of the normals of any two unit points in the neighborhood and perform an inverse cosine operation to obtain the minimum angle between the point normals, until the minimum angle between the maximum point normals is not less than 60° or the number of neighborhood points reaches 500, thus forming a neighborhood point set. Calculate the number of overlapping points between different neighborhood point sets, divide the number of overlapping points by the smaller number of points in the two neighborhood point sets to obtain the neighborhood overlap ratio, and pair and merge neighborhood point sets with an overlap ratio of not less than 80% according to the overlap ratio from large to small. The same neighborhood point set only participates in the merge once, and the neighborhood point sets that do not participate in the merge are kept separately to form local pipe wall point groups. For each local pipe wall point group, multiply the normal of each unit point by its own transpose to obtain a 3×3 point normal cross product matrix. Add up all the point normal cross product matrices in the group and divide by the number of pipe wall points to obtain the point normal direction matrix. The point normal direction matrix is decomposed into eigenvalues to determine the smallest eigenvalue among the three eigenvalues. The corresponding eigenvector is then normalized and determined as the local pipe extension direction. The X, Y, and Z coordinates of the pipe wall points in each local pipe wall point group are summed and divided by the number of pipe wall points in the group to obtain the center of the local pipe wall point group. The local pipe wall point group centers are paired from near to far. When the dot product of the two local pipe extension directions is less than 0, the three components of the extension direction of the latter local pipe are multiplied by -1. The pipe section is set with the center of the local pipe wall point group as the section setting position and the corresponding local pipe extension direction as the section normal. Project the pipe wall points in each local pipe wall point group onto the corresponding pipe cross-section. Determine the cross-section center according to the spatial distribution of the projected points. Encapsulate the pipe wall points, projected points, corresponding point normals, cross-section center, and cross-section normals into a pipe cross-section, where: Determine the center of the cross section, specifically as follows: For each pipe section, with the section setting position as the origin of the coordinate system, select the coordinate axis with the smallest absolute value of the dot product with the section normal from the X-axis, Y-axis and Z-axis as the reference axis. Take the cross product of the reference axis and the section normal and normalize it to obtain the first section coordinate axis. Then take the cross product of the section normal and the first section coordinate axis to obtain the second section coordinate axis. Calculate the displacement vector of each projection point relative to the cross-section setting position, and take the dot product of the displacement vector with the coordinate axes of the first and second cross-sections respectively to obtain the two-dimensional cross-section coordinates of each projection point; For each two-dimensional cross-sectional coordinate, the sum of the squares of the abscissa and ordinate and the negative value are used as the fitting term. The coefficient term is composed of the abscissa, ordinate and constant 1. The least squares method is used to calculate the fitting coefficients of the three circles. Divide the first two circle fitting coefficients by -2 respectively to obtain the coordinate components of the cross-section center on the first cross-section coordinate axis and the second cross-section coordinate axis. Multiply the two coordinate components by the corresponding cross-section coordinate axis and add them to the cross-section setting position to obtain the cross-section center. Using each pipe section as the current section, the center of the adjacent pipe sections is projected onto the local pipe extension direction of the current section. Based on the projection position, the preceding and succeeding sections are determined, generating a pipe section sequence, where: The pipeline cross-section sequence is generated as follows: For each pipe section, calculate the three-dimensional Euclidean distance between the center of the current section and the centers of the other pipe sections, arrange them in ascending order of distance, and select the 8 pipe sections with the closest distance as the neighboring pipe sections; Calculate the center displacement vector between the center of each adjacent pipe section and the center of the current section, and take the dot product of the center displacement vector and the current section normal to obtain the corresponding projection position; When the absolute value of the projected position is not less than 10% of the length of the center displacement vector, it is assigned to either the predecessor candidate set or the successor candidate set according to the sign of the projected position. When the absolute value of the projected position is less than 10% of the length of the center displacement vector, the center displacement vector is normalized and a dot product is taken with the cross-sectional normal of the corresponding adjacent pipe cross-section. When the absolute value of the dot product is not less than 0.866, it is included in the predecessor candidate set or successor candidate set according to the sign of the dot product. Other adjacent pipe cross-sections are not retained. Normalize the center displacement vector of each candidate section in the predecessor candidate set and the successor candidate set respectively, calculate the angle between different center displacement directions, and group the candidate sections with an angle of no more than 20° into the same direction group; Within each directional group, the pipe sections are arranged in ascending order of the three-dimensional Euclidean distance between the current section center and the candidate section center. The pipe section with the smallest distance is determined as the predecessor or successor section in the direction corresponding to the current section. Record the current cross section and its corresponding predecessor and successor cross sections to generate a pipeline cross section sequence; Connect the preceding and succeeding sections according to the pipe section sequence. For pipe sections with multiple preceding or succeeding sections, retain each connection branch and record the connection sequence of the pipe sections to form a section chain.
[0026] In this embodiment, generating the oil and gas pipeline reconstruction sample includes: Calculate the center-to-center distance between adjacent pipe sections along the cross-section chain. Compare each center-to-center distance with the median of the center-to-center distances between adjacent pipe sections in the cross-section chain. Mark the point cloud gaps between adjacent pipe sections whose center-to-center distance is greater than twice the median. Mark the gaps in the point cloud, specifically: Based on the connection relationship between the predecessor and successor sections in the cross-section chain, only one set is retained for repeated connections. The cross-section center of each set of adjacent pipe sections is extracted, and the three-dimensional Euclidean distance between the two cross-section centers is calculated to obtain the cross-section center distance. Arrange all the center distances of the cross sections in the cross section chain from smallest to largest. When the number of center distances is odd, take the value of the middle position. When the number is even, take the average of the values of the two middle positions to obtain the median of the center distances of the cross sections. The center distance of each cross section is compared with twice the median center distance of the cross sections. The adjacent pipe cross sections with a center distance greater than twice the median center distance of the cross sections are determined as the adjacent cross sections of the gap. Mark the corresponding connection section between the center of the cross section of the adjacent cross section as the point cloud gap, and record the connection relationship between the corresponding predecessor cross section and the successor cross section; The number of adjacent pipe sections connected to each pipe section is counted along the cross-section chain. Pipe sections connecting three or more adjacent pipe sections and their adjacent pipe wall areas are marked as branch intersection areas, where: Mark the intersection area of the branch roads, specifically: Based on the connection relationship between the predecessor and successor sections in the cross-section chain, the number of predecessor sections and successor sections of each pipe section are counted respectively, and the two are added together to obtain the number of adjacent pipe sections of the corresponding pipe section. Pipe sections with at least 3 adjacent pipe sections are defined as branch intersection sections, and the preceding and succeeding sections directly connected to the branch intersection sections are extracted. The pipe wall points contained in the branch intersection section and the preceding and succeeding sections directly connected to it are merged to form the branch intersection pipe wall area; The intersection section of the branch road and the corresponding pipe wall area of the branch road intersection are marked as the branch road intersection area; According to the connection sequence of the cross-section chain, the point cloud gaps and branch intersection regions are associated with the corresponding pipe cross-sections, and the two boundary cross-sections of the point cloud gaps and the connection cross-sections of the branch intersection regions are marked, wherein: Mark the two boundary sections and the connecting sections, specifically: For each point cloud gap, based on the connection relationship between the corresponding predecessor and successor sections, the predecessor and successor sections located on both sides of the point cloud gap and directly connected to the point cloud gap are marked as the two side boundary sections. For each branch intersection area, the corresponding branch intersection section is determined, and the predecessor and successor sections directly connected to the branch intersection section are extracted according to the connection sequence of the section chain. According to the connecting branches to which each preceding and succeeding section belongs, associate them with the corresponding branch intersection area and mark them as the connecting sections of the corresponding branch intersection area; The cross-section chain, point cloud gap, branch intersection area, two side boundary sections and connecting sections are encapsulated to generate oil and gas pipeline reconstruction samples.
[0027] In this embodiment, generating a complete point cloud of oil and gas pipelines includes: An improved PF-Net network is constructed by adding pipeline state reinjection units, gap recursive anchoring units, and pipe wall classification generation units to the PF-Net network. Furthermore, cross-section opening and closing migration units are embedded within the pipe wall classification generation units. The following steps are taken to build an improved PF-Net network: In the traditional PF-Net network, the multi-resolution encoder and the point pyramid decoding structure generated step by step from the primary center point, secondary center point to the minutiae point are retained. A pipeline state reinjection unit is added at the output of the multi-resolution encoder to fuse the cross-section center, cross-section normal, projection point, point cloud gap and branch intersection area in the oil and gas pipeline reconstruction sample with the corresponding coding features. A new gap recursive anchoring unit is added between the pipeline state reinjection unit and the point pyramid decoding structure. The gap anchor points are recursively generated and screened along the two boundary sections of the point cloud gap, and the anchoring features are input into the pipe wall hierarchical generation unit. Based on the traditional point pyramid decoding structure, a pipe wall hierarchical generation unit is constructed. The generation method of expanding detail points step by step according to the center point is changed to generating pipe wall points step by step according to the cross section center and the corresponding projection point. A cross section opening and closing migration unit is embedded in the pipe wall hierarchical generation unit to transfer the cross section center and cross section normal of each connecting cross section in the branch intersection area to the generation process of pipe wall points at each level along the corresponding connecting branch, thus obtaining an improved PF-Net network. The improved PF-Net network is trained using a joint loss consisting of pipe wall reconstruction error, notch anchoring closure error, cross-section opening / closing state transition error, and pipe wall point retention error. The trainable parameters of the improved PF-Net network are updated in reverse. The improved PF-Net network is considered to have completed convergence training when the absolute value of the difference between the joint losses of two consecutive validation sets is less than 0.0005 for six consecutive rounds. The improved PF-Net network is trained as follows: Point cloud samples of oil and gas pipelines with complete pipe walls are divided into training and validation sets in an 8:2 ratio. Using the complete pipe wall point cloud as the reference point cloud, pipe wall points corresponding to connecting branches in the intersection area of 2 to 6 consecutive pipe sections and branches are deleted along the cross-section chain to form the training input point cloud. The deleted pipe wall points are used as the real missing pipe wall points. The boundary sections on both sides of the gap are marked according to the deletion range. The connecting sections and the opening and closing states of the sections are marked according to the connection relationship of the complete pipe wall point cloud. Along each connecting branch, according to the connection sequence, the first unconnected section after the previous unconnected section and the current section is connected, the current section is connected to both the previous and previous sections, the current section is connected and the next section is not connected, and the first unconnected section after the exit state are marked as entering, holding, exiting, and closing states, respectively. Oil and gas pipeline reconstruction samples are constructed based on the training input point cloud and input into the improved PF-Net network to obtain the predicted probabilities of missing pipe wall points, gap anchoring chains, and the opening and closing states of each connecting branch. The diagonal length of the reference point cloud bounding box is calculated as the distance normalization scale. The three-dimensional Euclidean distance from each predicted missing pipe wall point to the nearest real missing pipe wall point and from each real missing pipe wall point to the nearest predicted missing pipe wall point are calculated separately. The two sets of distances are divided by the distance normalization scale, squared, and averaged. The two averages are then added together to obtain the pipe wall reconstruction error. The distance from the anchor points at both ends of the gap anchor chain to the center of the corresponding two boundary sections is divided by the distance between the center of the two boundary sections. The minimum angle between the extension direction of the first and last segments of the anchor chain and the normal of the corresponding boundary section is divided by 90°. The average of the four normalization results is taken to obtain the gap anchoring closure error. The cross-entropy of the four types of cross-section opening and closing state probabilities and the corresponding training labels is calculated to obtain the cross-section opening and closing state migration error. The predicted missing pipe wall points are merged with the training input point cloud. The average distance difference between the nearest 20 neighboring points (excluding itself) before and after the merging of each retained pipe wall point is calculated. The absolute value of this difference is divided by the distance normalization scale and averaged to obtain the pipe wall point preservation error. The joint loss is obtained by multiplying the four errors by 0.40, 0.25, 0.20 and 0.15 respectively and summing them. The Adam optimizer is used with an initial learning rate of 0.001. Every 16 training samples form a batch. The gradient is calculated based on the joint loss and the trainable parameters are updated in reverse. The joint loss of the validation set is calculated after each training round. When the absolute value of the difference between the joint losses of the validation sets of two adjacent rounds is less than 0.0005 for 6 consecutive rounds, the improved PF-Net network is considered to have completed convergence training. The reconstructed oil and gas pipeline samples are input into the pipeline state reinjection unit. The spatial distribution of the cross-section center, cross-section normal, and projection points of each pipeline cross-section is aggregated along the cross-section chain. The center displacement, cross-section normal rotation angle, and radial scale change of adjacent pipeline cross-sections are calculated. The calculation results are then backfilled to the corresponding pipeline cross-sections, retaining the corresponding pipe wall points. Point cloud gaps and branch intersection areas are correlated, and a pipeline state feature sequence is output, including: The pipeline-based reinjection unit includes: Cross-section state register: stores the cross-section center, cross-section normal, and projection point; Adjacent Section Difference Calculator: Calculates center displacement and section normal rotation angle; Radial Scale Calculator: Calculates radial scale and its variations; Status backfill buffer: The calculation results are backfilled to the corresponding pipe section; Pipe wall point retention index table: associates and retains the corresponding pipe wall points for each pipe section; Gap and Branch Association Queue: Associates point cloud gaps and branch intersection areas, and outputs a pipeline state feature sequence; State fusion mapper: aggregates pipeline state features and feeds them back into PF-Net encoded features; The specific process for the pipeline-based reinjection unit is as follows: The cross-section state register receives oil and gas pipeline reconstruction samples. Assume that the cross-section chain contains N pipeline cross-sections and B connection branches. The i-th pipeline cross-section contains Pi projection points. Write the 3D cross-section center, 3D cross-section normal and Pi×3D projection point coordinates along each connection branch according to the connection order. The adjacent section difference calculator pairs the predecessor and successor sections in each connecting branch, subtracts the center coordinates of the predecessor section from the center coordinates of the successor section to obtain the 3D center displacement, calculates the inverse cosine of the absolute value of the normal dot product of the two unit sections to obtain the normal rotation angle of the section from 0° to 90°, and associates it with the corresponding successor section. The radial scale calculator calculates the three-dimensional Euclidean distance from each projection point to the center of the corresponding section. Pi radial distances are sorted and the median is taken as the current radial scale. The radial scale of the successor section is subtracted from the radial scale of the predecessor section along the connecting branch and divided by the radial scale of the predecessor section to obtain the radial scale change. The state backfill buffer concatenates the 3D section center, 3D section normal, 3D center displacement, 1D section normal rotation, 1D current radial scale, and 1D radial scale change into a 12D section state vector, and backfills it to the corresponding pipe section. For pipe sections without a predecessor section in the connecting branch, the corresponding center displacement, section normal rotation, and radial scale change are set to 0. The pipe wall point preservation index table establishes the index relationship between the pipe section and the corresponding pipe wall point, associates the 12-dimensional section state vector with the corresponding pipe wall point, preserves the three-dimensional coordinates and point normal of the pipe wall point, and forms the pipe wall point index. The gap branch association queue writes 1D gap markers to the two side boundary sections of the point cloud gap, and 1D branch markers to the branch intersection sections and connecting sections. The remaining corresponding markers are set to 0. The two markers are concatenated with the 12D section state vector to form a 14D pipeline state feature vector. The association index between the two side boundary sections and the corresponding point cloud gaps and the association index between the connecting sections and the corresponding branch intersection areas are established. Suppose that the b-th connecting branch contains Nb pipeline sections. Nb×14-dimensional state subsequences are formed according to the connection order. The B state subsequences are summarized according to the connecting branch number and the pipe wall point index, point cloud gap association index and branch intersection area association index are added to output the pipeline state feature sequence. The state fusion mapper receives the pipeline state feature sequence and the 1920-dimensional encoding vector output by the PF-Net multi-resolution encoder. It maps each 14-dimensional pipeline state feature vector into a 1920-dimensional state mapping vector through a 14×1920-dimensional weight matrix and a 1920-dimensional bias vector. It averages the state mapping vector in each connection branch dimension by dimension to obtain the corresponding 1920-dimensional branch state vector. It arranges the B branch state vectors into a B×1920-dimensional branch state matrix. Then, it takes the maximum value in each dimension along the branch to obtain the 1920-dimensional pipeline state aggregation vector. It adds the aggregated vector to the 1920-dimensional encoding vector dimension by dimension to obtain the state backfeed encoding feature and transmits it to the gap recursive anchoring unit. The parameter configuration of the pipeline state reinjection unit includes the number of pipeline cross sections N, the number of connecting branches B, the number of projection points Pi, the number of pipeline cross sections Nb of each connecting branch, the dimension of the cross section state vector, the dimension of the pipeline state feature vector, and the dimension of the state mapping. The dimension of the cross section state vector is 12-dimensional, the dimension of the pipeline state feature vector is 14-dimensional, the dimension of the state mapping is 1920-dimensional, the dimension of the weight matrix of the state fusion mapper is 14×1920, the dimension of the bias vector is 1920, the range of the cross section normal rotation angle is 0° to 90°, and the gap mark and branch mark are both represented by 0 and 1 to indicate unassociated and associated, respectively. The pipeline state feature sequence is input into the gap recursive anchoring unit. Candidate anchor points are extrapolated from the two boundary sections of the point cloud gap towards the inside of the gap. Unclosed candidate branches are eliminated based on the center displacement, cross-sectional normal rotation angle, and radial scale changes. The remaining candidate anchor points are recursively extrapolated and connected to form a gap anchoring chain. The gap anchoring chain is backfilled into the pipeline state feature sequence, and the gap anchoring feature sequence is output, where: The gap recursive anchoring unit includes: Gap Boundary Index Table: Locating point cloud gaps and corresponding boundary sections on both sides; Two-way anchor point extrapolator: Generates candidate anchor points from the two side boundary sections into the point cloud gap; State recursive register: transmits center displacement, cross-sectional normal rotation angle, and radial dimension changes; Candidate closure checker: checks the center position, cross-sectional normal, and radial dimensions of the candidate anchor points on both sides; Candidate branch filtering queue: Filter candidate branches according to closure score and retain candidate anchor points; Recursive anchor buffer: receives and retains candidate anchors level by level and performs recursive extrapolation; Anchor chain connector: Connects candidate anchor points according to the recursive hierarchy to form a notched anchor chain; Anchoring Feature Backfiller: Backfills the gap anchoring chain into the pipe state feature sequence and outputs the gap anchoring feature sequence; The processing procedure for the recursive anchoring unit with the gap is as follows: The gap boundary index table receives the pipeline state feature sequence, locates the two side boundary sections of each point cloud gap according to the point cloud gap association index, extracts the corresponding 14-dimensional pipeline state feature vector, and obtains the state records of the three nearest non-gap pipeline sections facing the outside of the point cloud gap along the connecting branch. The state recursion register extracts the 3D center displacement, 1D cross-sectional normal rotation angle, and 1D radial scale change from the state records on both sides. It calculates the direction vector of each boundary cross-section pointing to the other side boundary cross-section. When the dot product of the center displacement and the corresponding direction vector is less than 0, the center displacement is reversed. When the recursion direction is opposite to the original predecessor-successor direction, the radial scale change is negative and divided by 1 and the sum of the original radial scale change to form 3 sets of recursion parameters on each side. The center distance between the two boundary cross-sections is used as the initial gap distance. The median of the 6 sets of center displacement moduli is taken as the anchor point recursion step size. The initial gap distance is divided by twice the anchor point recursion step size and rounded up to obtain the maximum recursion level R. The bidirectional anchor point extrapolator uses the two boundary sections as initial anchor points and generates three candidate anchor points according to three sets of recursive parameters on each side. Using the center displacement modulus as the extrapolation distance, the normal of the current anchor point is cross-producted with the normal of the boundary section on the other side. When the cross-product modulus is greater than 0, the cross-product result is normalized to obtain the normal rotation axis, and the normal of the current anchor point is rotated according to the normal rotation angle of the section. When the cross-product modulus is 0, the normal of the current anchor point is kept, and the normal of the candidate anchor point is obtained. The normal of the current anchor point and the normal of the candidate anchor point are added together, normalized, multiplied by the extrapolation distance, and then added to the center of the current anchor point to obtain the center of the candidate anchor point. The radial scale of the candidate anchor point is obtained by multiplying the current radial scale by 1 and the sum of the radial scale changes. The candidate closure checker pairs the candidate anchor points on both sides, divides the center distance by the initial gap distance to obtain the center displacement closure residual, divides the minimum normal angle by 90° to obtain the cross-sectional normal rotation angle closure residual, and divides the absolute value of the radial scale difference by the larger radial scale to obtain the radial scale change closure residual. These are weighted and accumulated by 0.50, 0.30, and 0.20 respectively to obtain the candidate branch closure score. The candidate branch selection queue is arranged in ascending order of closure score, and the two candidate branches with the lowest scores are retained at each level; The recursive anchoring buffer records the candidate branch number, parent branch number, candidate anchor points on both sides, and recursion level number. It then rewrites the recursion parameters into the status recursion register to continue extrapolation. When there is a branch among the retained candidate branches whose center distance between the anchor points on both sides is not greater than the anchor point recursion step size, the branch with the smallest closure score is selected. The average value of the center of the anchor points on both sides and the radial scale is taken respectively, and the normals of the anchor points on both sides are added together and normalized to obtain the closed anchor point and stop recursion. When the maximum recursion level R is reached and the anchor point is still not closed, the candidate branch with the smallest closure score is selected to generate the closed anchor point. The anchoring chain connector traces back layer by layer from the final anchoring branch along the parent branch number, arranging the anchor points on one side in ascending order of recursive layer number and the anchor points on the other side in descending order of recursive layer number. It then connects the anchor points with the boundary sections on both sides and the closed anchor points in sequence to form a gap anchoring chain. For the newly added anchor points, it calculates the 3D center displacement, 1D section normal rotation angle, and 1D radial scale change. It then concatenates these with the 3D anchor point center, 3D anchor point normal, and 1D radial scale to form a 12D anchor point state vector. Finally, it writes the larger value of the 1D gap marker and the branch markers of the boundary sections on both sides to form a 14D anchor point feature vector and updates the point cloud gap association index. The anchoring feature backfiller inserts the 14-dimensional anchor point feature vector between the corresponding two boundary sections. Assuming that the b-th connection branch originally contains Nb pipe sections and adds Mb anchor points, the Nb×14-dimensional state subsequence is converted into a (Nb+Mb)×14-dimensional gap anchoring state subsequence. Then, it is summarized according to the connection branch number, inheriting the pipe wall point index, point cloud gap association index, and branch intersection area association index, and associating it with the 1920-dimensional state backfill coding feature transmitted by the pipe state backfill unit, and outputting the gap anchoring feature sequence. Among them, the number of recursive states and candidate anchor points on each side are 3, the number of candidate branches retained in each layer is 2, the maximum number of recursive layers is R, and the weights of the three types of closed residuals are 0.50, 0.30 and 0.20, respectively. The notch anchoring feature sequence is input into the cross-section opening and closing migration unit in the pipe wall hierarchical generation unit. The cross-section states at the intersection of the cross-section chain and the notch anchoring chain migration branches are sequentially marked as opening entry state, opening hold state, and opening exit state. The remaining pipe cross-sections are marked as closed states. The circumferential generation interval is divided according to the relative position of the cross-section centers and the cross-section normal between connecting cross-sections. The cross-section states and circumferential generation intervals are then backfilled into the notch anchoring feature sequence, and the cross-section migration feature sequence is output, where: The cross-sectional opening and closing migration element includes: Branch Section Index Table: Locates the intersection area of branches and their corresponding connection sections; Section state migrater: migrates section state along the section chain and notch anchoring chain; Open / closed status marker: Marks the open state, open state, open state, and closed state; Connection section matcher: The connection section within the intersection area of associated branches; Circumferential datum calculator: Calculates the relative position of the cross-section center and the cross-section normal of the connecting section; Circumferential Interval Divider: Divides circumferentially generated intervals based on a circumferential reference. Migration Feature Backfiller: Backfills the cross-sectional state and circumferential generation interval, outputting a cross-sectional migration feature sequence; The processing procedure for the cross-section opening and closing migration element is as follows: The branch section index table receives the gap anchoring feature sequence, locates each branch intersection area, branch intersection section and connection section according to the branch intersection area association index, and extracts a 14-dimensional gap anchoring state subsequence containing the original pipeline section and the newly added anchor point according to the connection branch number; The cross-section state migrater arranges the pipe cross-sections and new anchor points along the cross-section chains and gap anchoring chains of each connecting branch, marks the branch intersection cross-sections, connecting cross-sections, and the pipe cross-sections and new anchor points between them as connected records, and marks the rest as unconnected records, forming a branch connection sequence according to the connection order; The open / closed state marker maps the 14×4 state weight matrix and 4-dimensional bias vector determined by the 14-dimensional gap anchor feature vectors after convergent training to 4-dimensional state values. After Softmax normalization, four types of state probabilities are obtained. The branch connectivity sequence is compared item by item. When the previous record is not connected and the current record is connected, it is marked as an open entry state. When the current record and the records before and after it are connected, it is marked as an open hold state. When the current record is connected and the next record is not connected, it is marked as an open exit state. The remaining records are marked as closed states. Adjacent records that are missing at the beginning or end of the sequence are treated as not connected, and the 4-dimensional state probabilities are retained as cross-sectional state codes. The connection section pairer associates each connection section according to the branch intersection area number. The 3D connection direction vector is obtained by subtracting the branch intersection section center from the connection section center. The connection direction vector and the projection of the connection section normal onto the branch intersection section normal are removed to obtain two circumferential components. When the dot product of the two circumferential components is less than 0, the next circumferential component is reversed. The non-zero circumferential components are then merged and normalized to obtain the connection circumferential pointing vector. When both circumferential components are zero, it is marked as an axial connection section. The circumferential reference calculator uses the center of the branch intersection section as the origin of the coordinate system. It selects the coordinate axis with the smallest absolute value of the dot product with the normal of the branch intersection section from the X-axis, Y-axis, and Z-axis as the reference axis. The reference axis is cross-producted with the normal of the branch intersection section and normalized to obtain the first circumferential axis. Then, the cross-product of the normal of the branch intersection section and the first circumferential axis is cross-producted and normalized to obtain the second circumferential axis. The dot product of the connecting circumferential pointing vector of the non-axial connecting section is calculated with the first and second circumferential axes respectively. The second projection component and the first projection component are subjected to a two-parameter arctangent operation. When the result is less than 0°, it is increased by 360° to obtain the connecting circumferential angle, which is then arranged in ascending order of angle. The circumferential interval divider calculates the angular interval between adjacent connecting circumferential angles along the positive direction from 0° to 360°, takes the midpoint of each angular interval as the circumferential dividing angle, and calculates the dividing angle of the first and last connecting circumferential angles according to the span of 360°. The adjacent circumferential dividing angles are used as the start and end angles of the circumferential generation interval of the corresponding connecting section. When there is only one connecting section participating in the circumferential interval division, the circumferential generation interval is taken as 90° before and after the corresponding connecting circumferential angle. The migration feature backfiller backfills the 4D cross-sectional state codes of each pipe section and newly added anchor point into the 14D gap anchoring feature vector according to the connection branch number. It also inherits the 2D circumferential interval start and end angles of the corresponding connection section along the connection branch. The two interval angles of the closed state and axial connection section are set to 0, forming a 20D cross-sectional migration feature vector. If the b-th connection branch contains Nb original pipe sections and Mb newly added anchor points, then a (Nb+Mb)×20-dimensional cross-sectional migration state subsequence is formed. Then, it is summarized according to the connection branch number, inheriting the pipe wall point index, point cloud gap association index, branch intersection area association index and 1920-dimensional state backfill coding features, and outputs the cross-sectional migration feature sequence. Among them, the cross-sectional state is represented by a 4-dimensional state probability, the state weight matrix has a dimension of 14×4, the bias vector has a dimension of 4, and the circumferential generation interval is represented by a 2-dimensional start and end angle. The cross-section migration feature sequence is input into the pipe wall hierarchical generation unit. The parent node of the cross-section is generated using the anchor points in the notch anchoring chain. The pipe wall sub-points are then expanded level by level from the parent node according to the cross-section state and circumferential generation interval. These sub-points are then merged with the pipe wall points retained in the cross-section migration feature sequence to generate a complete point cloud of the oil and gas pipeline. The pipe wall grading unit includes: Section parent node builder: Generates section parent nodes based on anchor points in the notch anchor chain; Generate a state index table: associate the cross-sectional state with the circumferential generation range; Hierarchical pipe wall expander: Pipe wall sub-points are generated step by step from the parent node of the cross section; Circumferential landing point controller: Generates the circumferential position of the pipe wall sub-points according to the circumferential generation interval; Pipe wall sub-point buffer: collects the final pipe wall sub-points; Retained Pipe Wall Point Index Table: Extracts retained pipe wall points from the cross-sectional migration feature sequence; Point cloud merger: merges pipe wall sub-points and retains pipe wall points to generate a complete point cloud of oil and gas pipelines; The processing steps of the pipe wall grading unit are as follows: The cross-section parent node builder receives the cross-section migration feature sequence, locates the new anchor point in the gap anchoring chain according to the point cloud gap association index, takes the corresponding 20-dimensional cross-section migration feature vector as the cross-section parent node feature, and extracts the 3-dimensional anchor point center, 3-dimensional anchor point normal and 1-dimensional radial scale. It is mapped to the 1920-dimensional parent node feature through a 20×1920-dimensional weight matrix and a 1920-dimensional bias vector. It is then added dimension by dimension with the state reinjection coding feature to obtain the 1920-dimensional pipe wall generation feature. The generation state index table associates the cross-sectional state and circumferential generation interval according to the connection branch number. The closed state uses the circumferential range of 0° to 360°, the three types of open states use the corresponding circumferential generation interval, and the non-closed state uses the circumferential range of 0° to 360° when the start and end angles of the interval are both 0°. The hierarchical pipe wall expander maps the 1920-dimensional pipe wall generation features to 16×2-dimensional first-level generation parameters. Each group includes a 1-dimensional circumferential parameter and a 1-dimensional radial correction parameter, forming 16 first-level pipe wall sub-point parameters. The generation parameters of each group are concatenated with the pipe wall generation features to form 1922-dimensional features. Each group is mapped to generate 2×2-dimensional second-level generation parameters, forming 32 second-level pipe wall sub-point parameters. Then, 64 third-level pipe wall sub-point parameters are generated according to the same dimensional relationship. The pipe wall sub-point parameters of each level are then transmitted to the circumferential landing point controller. The circumferential landing point controller uses the anchor point center as the coordinate origin and the anchor point normal as the cross section normal. It selects the coordinate axis with the smallest absolute value of the dot product with the cross section normal from the X-axis, Y-axis, and Z-axis as the reference axis. After two cross products and normalization, the first cross section axis and the second cross section axis are obtained. The first-level circumferential parameters are mapped to the circumferential generation interval through the Sigmoid operation. When the termination angle is less than the starting angle, it is increased by 360°. The second-level and third-level circumferential parameters are corrected by the hyperbolic tangent operation and the previous level circumferential angle is corrected by 1 / 32 and 1 / 64 of the interval width, respectively. Finally, the circumferential angle exceeding 360° is subtracted by 360°. The radial correction parameters of each level are multiplied by 2.5%, 1.25%, and 0.625% of the radial scale after the hyperbolic tangent operation, and are successively superimposed to the radial scale of the cross section parent node to obtain the generation radius. The three-dimensional coordinates of the final pipe wall sub-point are calculated based on the generation radius, the final circumferential angle, and the two cross section axes. The radial vector from the center of the cross section parent node to the final pipe wall sub-point is normalized to obtain the point normal. The pipe wall sub-point buffer collects the three-dimensional coordinates and point normals of the final pipe wall sub-points according to the point cloud gap number, connection branch number and cross section parent node sequence, forming a generated pipe wall point set; The pipe wall point index table is used to extract pipe wall points and their normals that did not participate in the gap filling, forming a set of retained pipe wall points. The point cloud merging unit merges the generated pipe wall point set with the retained pipe wall point set, calculates the three-dimensional Euclidean distance from each retained pipe wall point to the nearest retained pipe wall point other than itself and takes the median, removes generated pipe wall points whose distance from any retained pipe wall point is not greater than half of the median, and encapsulates the remaining generated pipe wall points and retained pipe wall points according to three-dimensional coordinates and point normals to generate a complete point cloud of oil and gas pipelines; Among them, the cross-sectional parent node features are 20-dimensional, the parent node mapping dimension and the state backfeeding coding feature dimension are both 1920-dimensional, the first level forms 16 pipe wall sub-point parameters, the second and third levels are expanded at a ratio of 1:2, and each cross-sectional parent node finally generates 64 pipe wall sub-points; This invention structurally optimizes the traditional PF-Net network for point cloud reconstruction scenarios of oil and gas pipelines. It introduces pipeline state reinjection units, gap recursive anchoring units, cross-section opening and closing migration units, and pipe wall hierarchical generation units, so that the model can further combine the continuous cross-section, gap extension and branch intersection structure information of oil and gas pipelines on the basis of the original point cloud encoding and hierarchical generation. To address the problem that the geometric state of continuous pipe walls is difficult to fully represent, a pipe state recharge unit is introduced, which incorporates the changes in the cross-section center, cross-section normal, and radial dimensions into the coding features, thereby enhancing the model's ability to represent the axial extension trend and cross-sectional change law of the pipe. To address the lack of structural references within point cloud gaps, a recursive gap anchoring unit is introduced. A gap anchoring chain is recursively established from both sides of the gap inwards, enabling the missing pipe wall to be continuously restored along the spatial extension trend of the existing pipe wall, thereby improving the stability of filling long-distance gaps and curved pipe sections. To address the issue of confusion regarding the pipe wall generation range in branch intersection areas, a cross-section opening and closing migration unit is introduced. This unit combines the cross-section opening and closing state with the circumferential generation interval to distinguish the generation range of different connecting branches, thereby enhancing the model's ability to identify the branch opening location and intersection structure. During the pipe wall generation stage, a hierarchical pipe wall generation unit is introduced, with the notch anchor point as the parent node of the cross section and the pipe wall sub-points are expanded step by step. This makes the generation process subject to the cross section state, circumferential range and radial scale constraints at the same time, improving the spatial continuity and geometric consistency between the generated pipe wall and the original pipe wall. The improved PF-Net network not only retains the original multi-resolution point cloud encoding and point pyramid hierarchical generation capabilities, but also achieves deep integration between the cross-sectional state of oil and gas pipelines, gap extension relationships, branch intersection structures and pipe wall point generation processes, thereby improving the integrity of missing pipe wall completion and the consistency of three-dimensional structure in oil and gas pipeline point cloud reconstruction.
[0028] In this embodiment, the generation of the reconstructed oil and gas pipeline structure includes: The complete point cloud along the oil and gas pipeline is divided into local pipe wall point groups. The center position of each local pipe wall point group is calculated, and adjacent center positions are connected according to spatial adjacency to generate the center trajectory, where: The center trajectory is generated as follows: For each local pipe wall point group, the eigenvector corresponding to the minimum eigenvalue of the accumulated matrix of points within the group is taken as the local pipe extension direction. The projection plane is constructed with the mean coordinate of the pipe wall points within the group as the reference and the local pipe extension direction as the normal. The pipe wall points within the group are projected and circle fitted. The center of the fitted circle is then mapped back to three-dimensional space to obtain the corresponding center position. Calculate the three-dimensional Euclidean distance between each center position to obtain the center distance, and take the median of the minimum center distances corresponding to each center position as the center adjacency spacing; The centers whose center distance is no more than twice the center adjacency spacing and which are among the four nearest centers are defined as spatially adjacent centers; Connect adjacent centers in the space according to the center distance from smallest to largest. When the number of adjacent centers is not less than 3, retain each connection branch and arrange the center positions and connection relationships according to the spatial extension order to generate the center trajectory. Trajectory sampling points are set along the central trajectory. The trajectory tangent at each sampling point is used as the normal to construct the corresponding normal section. The complete point cloud of the oil and gas pipeline is projected onto the corresponding normal section. The axial sequence is determined according to the arrangement order of the trajectory sampling points, where: The axial sequence is determined as follows: The sampling interval is set along the central trajectory according to the median distance between adjacent central positions, and the trajectory sampling points are set sequentially along each connecting branch; For internal trajectory sampling points, the trajectory tangent is obtained by subtracting the previous trajectory sampling point from the next trajectory sampling point. The first and last trajectory sampling points adopt the connection direction of adjacent trajectory sampling points, and the trajectory tangent is normalized. A normal section is constructed with each trajectory sampling point as the center and the corresponding trajectory tangent as the normal. For each pipe wall point, the closest trajectory sampling point is determined as the corresponding trajectory sampling point. The dot product of the displacement vector of the pipe wall point relative to the trajectory sampling point and the trajectory tangent is used as the axial distance. The product of the axial distance and the trajectory tangent is subtracted from the coordinates of the pipe wall point to obtain the corresponding projection point. Axial sequence positions are assigned along each connecting branch according to the arrangement order of the trajectory sampling points, and the corresponding projection points, pipe wall points and normal sections are associated to obtain the axial sequence positions of each normal section; Using the trajectory sampling points of each normal section as the section center, the projection points within the normal section are arranged according to their circumferential positions around the section center. The axial sequence and circumferential sequence are combined to form an axial-circumferential pipe wall lattice, and adjacent pipe wall points are connected to form a pipe wall mesh, wherein: The formation of a pipe wall mesh is as follows: For each connecting branch, the coordinate axis with the smallest absolute value of the dot product of the normal and the normal of the first normal section is selected as the reference axis. The cross product of the reference axis and the normal of the normal section is normalized to obtain the first section axis. The cross product of the normal of the normal section and the first section axis is normalized to obtain the second section axis. For the remaining normal sections, the first section axis of the previous normal section is projected onto the current normal section and normalized to obtain the first section axis. The cross product of the normal of the current normal section and the first section axis is normalized to obtain the second section axis. The displacement vector of the projection point relative to the center of the cross section is projected onto the two cross section axes. A two-parameter arctangent operation is performed based on the two projection components. When the result is less than 0°, 360° is added to obtain the circumferential angle. The projection points are arranged in ascending order of circumferential angle and assigned circumferential sequence positions. These are combined with the axial sequence positions and corresponding pipe wall points to form an axial-circumferential pipe wall point lattice. For axially adjacent normal sections, the circumferential angle is divided by 360° to obtain the normalized circumferential position, and the pipe wall points are paired according to the principle of the shortest circumferential distance between the first and last ends to form an axial pairing relationship. Connect adjacent circumferential and end pipe wall points within the same normal section to form a circumferential closed edge. Connect pipe wall points according to axial pairing to form an axial connecting edge. Divide the enclosed area into triangular mesh surfaces and summarize them to form a pipe wall mesh. The branching locations of the central trajectory and the changes in mesh connections between adjacent normal sections are identified along the pipe wall mesh. At the corresponding locations, the pipe wall mesh and the central trajectory are simultaneously segmented to form pipe segments and corresponding central trajectory segments. The pipe wall mesh of each segment is then encapsulated with its corresponding central trajectory segment to generate a reconstructed oil and gas pipeline structure. The reconstruction of the oil and gas pipeline structure is generated, specifically as follows: The number of adjacent connections at each center position is counted along the center trajectory corresponding to the pipe wall grid. The center position with no less than 3 adjacent connections is determined as the center trajectory branch position, and the axial sequence corresponding to the nearest trajectory sampling point is used as the branch split sequence. The axial sequence of adjacent normal sections is compared, and the circumferentially continuous axial connection edges are merged into connection intervals. When the connection intervals split or merge, the corresponding axial sequence is determined as the position of mesh connection change. Using the axial sequence of the beginning and end of each connecting branch, the branch splitting sequence, and the position of the mesh connection change as the splitting sequence, the center trajectory and pipe wall mesh are split simultaneously to form pipe segments and corresponding center trajectory segments. The pipe wall mesh of each pipe segment and the corresponding center trajectory segment are encapsulated one by one to generate the reconstructed oil and gas pipeline structure.
[0029] In this embodiment, generating the three-dimensional topology model of the oil and gas pipeline includes: Based on the pipe wall mesh of each pipe segment in the reconstructed oil and gas pipeline structure, continuously distributed boundary nodes at the boundary of each pipe segment are extracted to form the port boundary. The port boundary center is calculated, and the port normal is determined according to the end extension direction of the corresponding center trajectory segment. The port boundary, port boundary center and port normal are combined to form the pipe segment connection port. Calculate the center distance between port boundaries, the included angle of port normals, and the difference in perimeter between port boundaries between different pipe segment connection ports. Based on these three calculation results, filter and pair corresponding port boundaries to generate port pairing records, where: Generate port pairing records, specifically as follows: The port normals of each pipe segment connection port are unified to the direction from the inside of the corresponding pipe segment to the outside of the port. The three-dimensional Euclidean distance between the port boundary centers of different pipe segment connection ports is calculated. The port normals are normalized, multiplied by a dot product, and inverse cosine is calculated to obtain the port normal angle. The perimeter of the port boundary is obtained by accumulating the distances between adjacent boundary nodes along the port boundary, and the absolute value of the difference between the two port boundary perimeters is calculated to obtain the port boundary perimeter difference. Divide the center distance of the port boundaries by the average of the perimeters of the two port boundaries, divide the difference between the port normal angle and 180° by 180°, and divide the difference in the perimeters of the port boundaries by the larger of the two port boundary perimeters to obtain the center distance difference, normal difference, and perimeter difference. Arrange the port pairs according to the three differences from smallest to largest, round up one-quarter of the total number of port pairs to obtain the number of screening pairs, and determine the port pairs in which all three differences are within the corresponding number of screening pairs as candidate port pairs; For each candidate port pair, a weighted sum is calculated based on the differences in center distance, normal direction, and perimeter (0.50, 0.30, and 0.20, respectively) to obtain the port matching value. The two pipe segments with the smallest port matching values are then identified as the paired ports. Associate two paired ports, their respective pipe segments, and port matching values to generate a port pairing record; Associating corresponding pipe segments with port pairing records, establishing adjacency connections for corresponding pipe segments whose port boundaries have been paired, and retaining the adjacency branches corresponding to the connecting ports of each pipe segment for cases where the same pipe segment corresponds to multiple port pairing records, thus forming pipe segment adjacency relationships; By associating the pipe wall mesh, pipe connection ports, port pairing records, and pipe adjacency relationships of each pipe segment, a three-dimensional topology model of the oil and gas pipeline is generated.
[0030] In this embodiment, generating the three-dimensional model of the oil and gas pipeline includes: Adjacent pipe segments are determined based on the adjacency relationship of pipe segments in the three-dimensional topology model of oil and gas pipelines, the pipe wall mesh corresponding to the adjacent pipe segments is extracted, and the boundary pipe wall between the pipe segment connection ports is located. Cut the junction pipe wall along the port boundary of the pipe segment connection port, extract the boundary nodes of the cut boundaries on both sides, and sort them according to the circumferential position of the boundary nodes around the center of the corresponding port boundary, where: The boundary nodes are sorted as follows: Cut the corresponding junction pipe wall along the port boundary of the two paired ports, extract the continuously connected boundary nodes after cutting, and form the two cutting boundaries respectively; Calculate the three-dimensional Euclidean distance between the two cutting boundary nodes, and take the pair of boundary nodes with the smallest distance as the starting nodes for circumferential sorting on both sides. The circumferential increasing direction is obtained by cross-product of the normal of one port with the direction of the starting node relative to the center of the port boundary. From the two adjacent boundary nodes of the starting node, the node along the circumferential increasing direction is selected as the circumferential increasing node. On the other side, the adjacent boundary node that is smaller in distance from the circumferential increasing node is selected as the corresponding circumferential increasing node. Arrange adjacent boundary nodes sequentially from the starting nodes on both sides along the corresponding circumferential increasing direction, and assign circumferential order positions according to the arrangement order around the center of the corresponding port boundary to complete the circumferential sorting of the cutting boundary nodes on both sides. Pair the corresponding boundary nodes on both sides of the clipping boundary according to the circumferential order. Insert transition nodes for clipping boundaries with inconsistent numbers of boundary nodes to form a clipping boundary sequence with consistent numbers of nodes on both sides, where: The cut boundary sequence is formed as follows: Count the number of boundary nodes on both sides of the clipping boundary. When the number of boundary nodes on both sides is inconsistent, calculate the three-dimensional Euclidean distance between adjacent boundary nodes in the circumferential direction in the clipping boundary with fewer nodes, select the adjacent node with the largest distance, and take the average of the two coordinates as the transition node to insert at the corresponding position. Rearrange the boundary nodes after the inserted transition nodes according to their circumferential positions and assign them circumferential order positions. Repeat the insertion operation until the number of boundary nodes on both sides is the same. The boundary nodes on both sides are re-paired according to the circumferential order and arranged in sequence to form a cutting boundary sequence with the same number of nodes on both sides; Connect corresponding nodes and adjacent nodes according to the node pairing relationship of the cut boundary sequence to construct a transition mesh that spans both cut boundaries. Then, merge the transition mesh with the pipe wall mesh of the adjacent pipe segment to generate a 3D model of the oil and gas pipeline, where: Generate a 3D model of the oil and gas pipeline, specifically as follows: According to the circumferential order of the cutting boundary sequence, connect the nodes with the same order on both sides to form a cross-boundary connection edge, and connect the adjacent circumferential nodes on each side to form a circumferential connection edge. According to the circumferential end-to-end relationship, the area enclosed by the pairing relationship of two adjacent sets of nodes is divided into triangular grid surfaces along the diagonal and connected in sequence to form a transition grid. Insert the transition node into the corresponding cut boundary edge of the adjacent pipe segment wall mesh and split the corresponding mesh edge. Merge the transition mesh with the pipe wall mesh of the adjacent pipe segment, merge duplicate nodes and duplicate mesh edges, and generate a 3D model of the oil and gas pipeline.
[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to a 3D modeling task of an oil and gas pipeline. A multi-site point cloud was collected for a section containing straight pipes, bends, diameter transition pipes, and branch intersections. A total of 9 scanning stations were set up, with an initial point cloud totaling 4.863 million points. The main pipe has an outer diameter of 508 mm, and the outer diameter of the transition area transitions from 508 mm to 406 mm. Due to occlusion and scanning angle limitations, continuous gaps exist on the inner side of the bends, the transition area, and the back side of the branches. The maximum axial length of the gap is 0.54 m, and there is a positional deviation of 10–20 mm between adjacent scanning stations.
[0032] The point cloud was denoised after being divided by scanning station. Taking the 4th scanning station as an example, the original number of points was 538,000. After calculating the spatial distance between each point and its neighbors, 19,200 outliers were removed, leaving 518,800 valid points. After denoising the 9 groups of point clouds, the total number of points was reduced from 4,863,000 to 4,697,000. Subsequently, the coordinate units and coordinate axis arrangement rules were standardized to maintain the original relative pose of the scanning stations. For each valid point, 24 neighboring points were selected to calculate the local covariance matrix. The eigenvector corresponding to the smallest eigenvalue was used as the point normal. The 3D coordinates, point normals, and scanning station identifiers were encapsulated to form a standard point cloud for oil and gas pipelines.
[0033] Taking the 4th and 5th scanning stations as examples, the two point clouds contain 518,800 and 493,600 valid points, respectively. Based on the spatial distance between points and the minimum angle between the point normals, 87,100 corresponding points were selected, of which 79,600 consecutive corresponding points formed overlapping pipe segments. Before pose correction, the average spatial distance was 15.8 mm, and the minimum angle between the point normals was 7.4°. The calculated inter-station rotation was 1.36°, and the translation was 18.7 mm. After correction, the average spatial distance decreased to 3.2 mm, and the minimum angle between the point normals decreased to 2.1°. After connecting and merging the overlapping points from the nine scanning stations, a continuous point cloud of oil and gas pipelines consisting of 3,274,000 points was formed.
[0034] Local pipe wall point groups were divided along the continuous point cloud of the oil and gas pipeline, with each group averaging approximately 3500 points. The local pipeline extension direction was determined based on the point normal, and pipeline cross-sections were established, resulting in a total of 782 pipeline cross-sections. One bend cross-section contained 3418 projected points, with an equivalent radius of 253.1 mm. After projecting the center of adjacent pipeline cross-sections, the preceding cross-section was located at -48.6 mm, and the subsequent cross-section at 49.2 mm, forming a cross-section chain. The median distance between the centers of adjacent cross-sections in the cross-section chain was 49.1 mm. In a branch intersection area, the distance between the centers of adjacent cross-sections was 128.7 mm, greater than 98.2 mm, and was marked as a point cloud gap. This pipeline cross-section connected one preceding cross-section and two subsequent cross-sections and was marked as a branch intersection area. After processing all cross-sections, a total of 6 point cloud gaps were obtained, and these were associated with the corresponding boundary cross-sections on both sides and connecting cross-sections to form a reconstructed oil and gas pipeline sample.
[0035] The improved PF-Net network was trained using 2400 point cloud samples, with 1680 sets used for training, 360 sets for validation, and 360 sets for testing. Each complete label set contained 32768 points, and simulated pipe wall defects at a ratio of 8%–30%. The training samples covered straight pipes, bends, reducers, and branch intersections. The joint loss was calculated using pipe wall reconstruction error, gap anchoring closure error, cross-section opening / closing state migration error, and pipe wall point retention error. The joint loss was 0.0148 in the 76th round of validation, 0.0145 in the 77th round, and 0.0142 in the 78th round. After that, the absolute value of the difference between the joint losses of adjacent rounds was less than 0.0005 for six consecutive rounds, indicating successful training.
[0036] One set of training samples containing branch intersection areas was selected for simulation. The complete label contained 32,768 points. After removing 6,214 pipe wall points, 26,554 points were retained. The gap length was 0.46m, and the missing rate was 18.96%. The pipeline state reinjection unit was spatially distributed along the cross-sectional chain aggregation cross-section center, cross-sectional normal, and projection point. The average center displacement of the continuous cross-sections on both sides of the gap was 48.3mm, and the cross-sectional normal rotation angles were 2.4°, 3.1°, 4.6°, and 3.8°, respectively. The radial scale variation ranged from -4.2mm to 3.5mm, and the corresponding results were backfilled to each pipeline cross-section.
[0037] The recursive anchoring unit for the gap is pushed outward from the boundary sections on both sides of the point cloud gap, initially forming 7 candidate branches. When recursing to the 4th layer, the center positions of 2 branches are offset by more than 12mm, and the normal rotation angle of the section of 1 branch reaches 11.6°, which is inconsistent with the trend of the boundary sections on both sides, and therefore they are eliminated. The remaining candidate branches continue recursion, and one of them closes at the 9th anchor point, with a center distance of 3.8mm and a normal difference of 1.7°, forming a gap anchoring chain consisting of 9 anchor points.
[0038] The branch intersection area contains 10 relevant pipe sections, of which 2 are marked as open-entry, 4 as open-maintained, 2 as open-exit, and the remaining 2 as closed. Based on the relative positions of the section centers and the section normals between the connecting sections, the circumferential generation range is determined to be 68°–151°. The pipe wall hierarchical generation unit generates section parent nodes with 9 anchor points, and expands 6468 candidate pipe wall sub-points step by step according to the section state and the circumferential generation range. After merging with the retained pipe wall points and removing 254 duplicate points, 6214 effective completion points are obtained, restoring the training sample to 32768 complete points.
[0039] After generating the complete point cloud of the oil and gas pipeline, the center positions of local pipe wall point groups are calculated and connected according to spatial adjacency to form a central trajectory. 764 trajectory sampling points are set along the central trajectory, and the complete point cloud of the oil and gas pipeline is projected onto the corresponding normal sections, rearranged according to axial and circumferential positions. The 326th normal section contains 3872 projected points. After connecting adjacent pipe wall points, approximately 186,000 triangular mesh surfaces are formed. Based on the branch positions of the central trajectory and the changes in mesh connections between adjacent normal sections, 14 pipe segments and corresponding central trajectory segments are obtained.
[0040] Thirty pipe segment connection ports were extracted from the boundaries of 14 pipe segments. Taking one candidate port pair as an example, the center distance between the port boundaries was 6.4 mm, the port normal angle was 176.8°, and the difference in the port boundary perimeter was 9.7 mm. After screening, port pairing records were formed. A total of 15 pairing records were obtained for all ports, and the adjacency relationship between the 14 pipe segments was established. The pipe wall at the junction of adjacent pipe segments was trimmed. One set of trimmed boundaries contained 84 and 79 boundary nodes on the two sides, respectively. Five transition nodes were inserted on the side with fewer nodes, so that both sides formed 84 ordered boundary nodes. Then, the corresponding nodes and adjacent nodes were connected to form a transition mesh, and finally, a 3D model of the oil and gas pipeline was generated.
[0041] On the same batch of 360 test samples, the present invention was compared with traditional methods that employ nearest-point registration, local surface interpolation, and direct mesh reconstruction.
[0042] Regarding the average registration residual in the overlapping area of multiple sites, the traditional method yields 7.6 mm, while this invention yields 3.3 mm. Regarding the average reconstruction error of the missing pipe wall, the traditional method is 12.1 mm, while the present invention is 4.5 mm; Regarding the average deviation of the center trajectory in the bend area, the traditional method yields 9.4 mm, while the present invention yields 3.8 mm. Regarding the average error of the cross-sectional radius in the variable diameter region, the traditional method results in 8.1 mm, while the present invention results in 3.0 mm. Regarding the average deviation of the branch opening boundary, the traditional method yields 17.9 mm, while the present invention yields 5.7 mm. Regarding the accuracy of pipe segment topology connections, the traditional method achieves 91.7%, while this invention achieves 98.9%. Regarding the final model mesh unclosed rate, the traditional method yields 6.4%, while this invention yields 1.1%. Regarding the average modeling time for a single point cloud sample, the traditional method takes 208 seconds, while this invention takes 193 seconds.
[0043] Further analysis of 360 test samples revealed that traditional methods generated 22 obviously erroneous branch connections, while this invention generated 4. For 68 test samples with a notch length greater than 0.4m, the average reconstruction error of the traditional method was 15.6mm, while that of this invention was 5.8mm. For 84 test samples of branch intersections, the traditional method achieved a branch structure integrity rate of 88.1%, while the present invention achieved 97.6%.
[0044] As can be seen from Example 1, the present invention can complete continuous point cloud splicing, missing pipe wall restoration, pipe segment topology establishment and boundary mesh reorganization under the conditions of positional deviation, continuous gaps and branch occlusion in multi-site pipe wall point clouds. It reduces the error of pipe wall misalignment and gap filling, improves the accuracy of branch opening restoration and pipe segment connection integrity, and verifies the feasibility of the method for generating three-dimensional models of complex oil and gas pipelines.
[0045] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for generating a three-dimensional model of an oil and gas pipeline based on point cloud reconstruction, characterized in that, include: Collect point clouds from multiple sites along oil and gas pipelines and perform preprocessing to generate a standard point cloud set for oil and gas pipelines. Based on the standard point cloud of oil and gas pipelines, the overlapping pipe segments of adjacent scanning stations are paired, the pose between stations is corrected and the connection is continued along the pipeline extension direction to generate a continuous point cloud of oil and gas pipelines. Divide the pipeline cross-section along the continuous point cloud of the oil and gas pipeline, connect adjacent cross-sections in sequence to form a cross-section chain, mark the point cloud gaps and branch intersection areas, and generate oil and gas pipeline reconstruction samples. The oil and gas pipeline reconstruction sample is input into the improved PF-Net network, which includes a pipeline state reinjection unit, a gap recursive anchoring unit, a pipe wall hierarchical generation unit, and a cross-section opening and closing migration unit to generate a complete point cloud of the oil and gas pipeline. Based on the extraction of the center trajectory from the complete point cloud of the oil and gas pipeline, the complete point cloud of the oil and gas pipeline is projected onto the normal section of the center trajectory and rearranged axially and circumferentially to construct the pipe wall mesh and divide the pipe segments, thereby generating the reconstructed structure of the oil and gas pipeline. Based on the reconstructed structure of the oil and gas pipeline, the pipe segment connection ports are extracted, the port boundaries are matched and the pipe segment adjacency relationship is established to generate a three-dimensional topology model of the oil and gas pipeline. Based on the 3D topology model of the oil and gas pipeline, the boundary wall of adjacent pipe segments is cut, the cut boundary is reorganized and a transition mesh is constructed to generate the 3D model of the oil and gas pipeline.
2. The method of claim 1, wherein, The point cloud of the oil and gas pipeline multi-site specifically includes the point cloud of straight pipe sections, the point cloud of bends, the point cloud of pipe sections with varying diameters, the point cloud of branch intersections, and the point cloud of pipe fitting connection areas.
3. The method of claim 1, wherein, The generated standard point cluster for oil and gas pipelines includes: Divide the oil and gas pipeline multi-site point cloud according to the scanning station, calculate the spatial distance between each point and its neighboring points and remove outliers to generate a denoised point cloud of the oil and gas pipeline. Unify the coordinate units and coordinate axis arrangement rules of the denoised point cloud of oil and gas pipelines, without changing the original relative pose between each scanning station, and generate a standard point cloud of oil and gas pipelines. For each point in the standard point cloud of oil and gas pipelines, neighborhood points are selected, and the local covariance matrix is calculated based on the coordinates of the neighborhood points. The eigenvector corresponding to the minimum eigenvalue is determined as the point normal. The three-dimensional coordinates, point normals, and scanning station identifiers of the standard point cloud of oil and gas pipelines are encapsulated to generate a standard point cloud set of oil and gas pipelines.
4. The method of claim 1, wherein, The generation of continuous point clouds for oil and gas pipelines includes: The standard point cloud of oil and gas pipelines is divided according to the scanning station identification. The point clouds of adjacent scanning stations are paired. Corresponding points are selected according to the spatial distance between points and the minimum angle between the point normals. The pipe wall area where the corresponding points are continuously distributed is determined as the overlapping pipe section. Based on the three-dimensional coordinates and point normals of corresponding points within the overlapping pipe sections, the rotation and translation between adjacent scanning stations are calculated. The point cloud pose of the corresponding scanning station is corrected according to the rotation and translation, and the normal of the corresponding point is corrected synchronously according to the rotation. According to the pairing order of adjacent scanning stations, the overlapping pipe segments after pose correction are used as the continuation area, and the non-overlapping pipe segments are spliced sequentially along the extension direction of the oil and gas pipeline to generate the oil and gas pipeline continuation point cloud. Merge the corresponding points and corrected point normals of overlapping pipe segments in the oil and gas pipeline connection point cloud, and retain the pipe wall points and corrected point normals of non-overlapping pipe segments to generate a continuous point cloud of oil and gas pipelines.
5. The method of claim 1, wherein, The formation of the cross-section chain includes: The continuous point cloud along the oil and gas pipeline is divided into local pipe wall point groups. The local pipeline extension direction is determined according to the point normal of each local pipe wall point group, and the pipeline section is set with the local pipeline extension direction as the section normal. Project the pipe wall points in each local pipe wall point group onto the corresponding pipe section, determine the section center according to the spatial distribution of the projection points, and encapsulate the pipe wall points, projection points, corresponding point normals, section center and section normals into a pipe section; Using each pipe section as the current section, the center of the adjacent pipe sections is projected onto the local pipe extension direction of the current section. Based on the projection position, the preceding and subsequent sections are determined to generate a pipe section sequence. Connect the preceding and succeeding sections according to the pipe section sequence. For pipe sections with multiple preceding or succeeding sections, retain each connection branch and record the connection sequence of the pipe sections to form a section chain.
6. The method of claim 1, wherein, The generated oil and gas pipeline reconstruction sample includes: Calculate the center-to-center distance between adjacent pipe sections along the cross-section chain, compare each center-to-center distance with the median of the center-to-center distances between adjacent pipe sections in the cross-section chain, and mark the point cloud gaps between adjacent pipe sections whose center-to-center distance is greater than twice the median. Count the number of adjacent pipe sections connected to each pipe section along the cross-section chain, and mark the pipe sections that connect three or more adjacent pipe sections and their adjacent pipe wall areas as branch intersection areas; According to the connection sequence of the cross-section chain, the point cloud gap and the branch intersection area are respectively associated with the corresponding pipe cross-section, and the two boundary cross-sections of the point cloud gap and the connection cross-section of the branch intersection area are marked. The cross-section chain, point cloud gap, branch intersection area, two side boundary sections and connecting sections are encapsulated to generate oil and gas pipeline reconstruction samples.
7. The method of claim 1, wherein, The generation of the complete point cloud of the oil and gas pipeline includes: An improved PF-Net network is constructed by adding a pipeline state reinjection unit, a gap recursive anchoring unit, and a pipe wall classification generation unit to the PF-Net network, and embedding a cross-section opening and closing migration unit in the pipe wall classification generation unit. The improved PF-Net network is trained by using the pipe wall reconstruction error, notch anchoring closure error, cross-section opening and closing state transition error and pipe wall point holding error as the joint loss. The trainable parameters of the improved PF-Net network are updated in reverse. When the absolute value of the difference between the joint loss of two adjacent rounds of validation sets is less than 0.0005 for 6 consecutive rounds, the improved PF-Net network is considered to have completed convergence training. The reconstructed samples of oil and gas pipelines are input into the pipeline status reinjection unit. The spatial distribution of the cross-section center, cross-section normal and projection points of each pipeline cross-section is aggregated along the cross-section chain. The center displacement, cross-section normal rotation angle and radial scale change of adjacent pipeline cross-sections are calculated. The calculation results are backfilled to the corresponding pipeline cross-sections, and the corresponding pipe wall points are retained. The point cloud gaps and branch intersection areas are associated, and the pipeline status feature sequence is output. Input the pipeline state feature sequence into the gap recursive anchoring unit, push candidate anchor points outward from the two boundary sections of the point cloud gap into the gap, eliminate unclosed candidate branches according to the center displacement, cross section normal rotation angle and radial scale change, recursively push out the remaining candidate anchor points and connect them to form a gap anchoring chain, backfill the gap anchoring chain into the pipeline state feature sequence, and output the gap anchoring feature sequence. The gap anchoring feature sequence is input into the cross-section opening and closing migration unit in the pipe wall hierarchical generation unit. The cross-section state of the intersection area of the cross-section chain and the gap anchoring chain migration branch is used to mark the corresponding pipe cross-sections as the opening entry state, the opening holding state, and the opening exit state in sequence. The remaining pipe cross-sections are marked as the closed state. The circumferential generation interval is divided according to the relative position of the cross-section center and the cross-section normal between the connecting cross-sections. The cross-section state and the circumferential generation interval are backfilled into the gap anchoring feature sequence, and the cross-section migration feature sequence is output. The cross-section migration feature sequence is input into the pipe wall hierarchical generation unit. The cross-section parent node is generated by the anchor point in the gap anchoring chain. The pipe wall sub-points are expanded step by step from the cross-section parent node according to the cross-section state and the circumferential generation interval. The pipe wall sub-points are merged with the pipe wall points retained in the cross-section migration feature sequence to generate a complete point cloud of the oil and gas pipeline.
8. The method of claim 1, wherein, The generated oil and gas pipeline structure reconstruction includes: Divide the complete point cloud of the oil and gas pipeline into local pipe wall point groups, calculate the center position of each local pipe wall point group, and connect adjacent center positions according to spatial adjacency to generate the center trajectory; Set up trajectory sampling points along the central trajectory, and construct the corresponding normal section using the trajectory tangent at each trajectory sampling point as the normal. Project the complete point cloud of the oil and gas pipeline onto the corresponding normal section, and determine the axial sequence according to the arrangement order of the trajectory sampling points. Using the trajectory sampling points of each normal section as the center of the section, the projection points in the normal section are arranged according to the circumferential position of the projection points around the center of the section. The axial sequence and the circumferential sequence are combined to form an axial-circumferential pipe wall lattice, and adjacent pipe wall points are connected to form a pipe wall grid. The branch positions of the central trajectory and the changes in the grid connection between adjacent normal sections are identified along the pipe wall grid. The pipe wall grid and the central trajectory are simultaneously segmented at the corresponding positions to form pipe segments and corresponding central trajectory segments. The pipe wall grid of each pipe segment and the corresponding central trajectory segment are encapsulated to generate the reconstructed structure of the oil and gas pipeline.
9. The method of claim 1, wherein, The generation of the three-dimensional topology model of the oil and gas pipeline includes: Based on the pipe wall mesh of each pipe segment in the reconstructed oil and gas pipeline structure, continuously distributed boundary nodes at the boundary of each pipe segment are extracted to form the port boundary. The port boundary center is calculated, and the port normal is determined according to the end extension direction of the corresponding center trajectory segment. The port boundary, port boundary center and port normal are combined to form the pipe segment connection port. Calculate the center distance between the port boundaries, the included angle of the port normal, and the difference in the perimeter of the port boundaries between the connection ports of different pipe segments. Based on the three calculation results, filter and pair the corresponding port boundaries to generate port pairing records. Associating corresponding pipe segments with port pairing records, establishing adjacency connections for corresponding pipe segments whose port boundaries have been paired, and retaining the adjacency branches corresponding to the connecting ports of each pipe segment for cases where the same pipe segment corresponds to multiple port pairing records, thus forming pipe segment adjacency relationships; By associating the pipe wall mesh, pipe connection ports, port pairing records, and pipe adjacency relationships of each pipe segment, a three-dimensional topology model of the oil and gas pipeline is generated.
10. The method for generating a three-dimensional model of an oil and gas pipeline based on point cloud reconstruction according to claim 1, characterized in that, The generation of the three-dimensional model of the oil and gas pipeline includes: Adjacent pipe segments are determined based on the adjacency relationship of pipe segments in the three-dimensional topology model of oil and gas pipelines, the pipe wall mesh corresponding to the adjacent pipe segments is extracted, and the boundary pipe wall between the pipe segment connection ports is located. Cut the junction pipe wall along the port boundary of the pipe segment connection port, extract the boundary nodes of the cut boundaries on both sides, and sort them according to the circumferential position of the boundary nodes around the center of the corresponding port boundary. Pair the corresponding boundary nodes in the two sides of the clipping boundary according to the circumferential order, and insert transition nodes for clipping boundaries with inconsistent numbers of boundary nodes to form a clipping boundary sequence with consistent numbers of nodes on both sides. Connect corresponding nodes and adjacent nodes according to the node pairing relationship of the cut boundary sequence, construct a transition mesh that spans both cut boundaries, and merge the transition mesh with the pipe wall mesh of the adjacent pipe segment to generate a three-dimensional model of the oil and gas pipeline.
