A Data-Driven Optimization Method for 3D Models of Marine Platforms

Abstract

This invention relates to the field of 3D modeling technology, and specifically discloses a data-driven method for optimizing 3D models of marine platforms. The method corrects the main model of the marine platform in the engineering rectangular coordinate system according to its actual tilt attitude and maps it to the geographic spherical coordinate system. By identifying the rotation transformation singularity through latitude and longitude jumps, local geometric features in the neighborhood of the singularity are extracted, an adaptive direction field is constructed and rays are emitted, and new mesh patches are constructed based on the ray intersections. Then, the correction result is inversely transformed back to the engineering rectangular coordinate system and superimposed on the seabed basic model to obtain a continuous 3D model with correct geometric relationships.

Description

Technical Field

[0001] This invention relates to the field of 3D modeling technology, and more specifically to a data-driven method for optimizing 3D models of marine platforms. Background Technology

[0002] In the 3D modeling of offshore platforms, the main body of the platform is typically modeled using an engineering rectangular coordinate system for detailed structural design, while the seabed foundation uses a geographic spherical coordinate system to match the curvature of the Earth. Current technologies, when fusing models from these two coordinate systems using a data-driven approach, usually only perform coordinate calibration at the boundary between the platform body and the seabed foundation before directly splicing them together, without considering the attitude and rotational transformations caused by factors such as tilted landing and leg penetration into the mud during actual installation. When the platform body is superimposed on the seabed foundation model in an tilted posture, the rotational transformation between the coordinate systems introduces mathematical singularities, resulting in implicit geometric distortions at the interface between the platform body and the seabed foundation. These distortions cannot be identified during routine modeling checks but will accumulate errors in subsequent structural simulation analysis and construction process simulation applications. Summary of the Invention

[0003] The purpose of this invention is to provide a data-driven method for optimizing three-dimensional models of marine platforms, thereby solving the following technical problems.

[0004] The objective of this invention can be achieved through the following technical solutions: A data-driven method for optimizing a 3D model of a marine platform includes the following steps: Obtain the main model of the marine platform and the basic model of the seabed. The main model of the marine platform is constructed using an engineering rectangular coordinate system, while the basic model of the seabed is constructed using a geographic spherical coordinate system. The vertex coordinates of the main model of the offshore platform are corrected to obtain corrected coordinates, and then the corrected coordinates are mapped to the geographic spherical coordinate system; Check whether the corrected coordinates jump during the mapping process, and mark the corrected coordinates that jump as rotation transformation singularities; Multiple rays extend outward from the rotational transformation singularity as the center, and new mesh patches are constructed based on the rays; The main body model of the marine platform is corrected based on the newly added mesh patches. The corrected main body model of the marine platform is then superimposed on the seabed base model to output the optimized three-dimensional model of the marine platform.

[0005] As a further aspect of the present invention: obtaining the correction coordinates includes: Read the modeling file of the main offshore platform model. The vertex coordinates stored in the modeling file correspond to the geometric data of the main offshore platform model in an upright position. In the upright position, the deck plane of the main offshore platform model is parallel to the horizontal plane, and the leg axis of the main offshore platform model is perpendicular to the horizontal plane. Use the vertex coordinates read from the modeling file as the initial vertex coordinates. Obtain the tilt attitude parameters of the main offshore platform model. The tilt attitude parameters include the rotation angles around the horizontal axis and the rotation angles around the vertical axis in the engineering rectangular coordinate system. Construct a rotation matrix based on the rotation angles. Multiply the initial vertex coordinates by the rotation matrix to obtain the intermediate vertex coordinates. Obtain the translation parameters of the main offshore platform model in the engineering rectangular coordinate system. Add the intermediate vertex coordinates to the translation parameters to obtain the tilt state vertex coordinates. Use the tilt state vertex coordinates as the correction coordinates.

[0006] As a further aspect of the present invention: checking for the existence of a transition includes: After mapping the calibration coordinates to the geographic spherical coordinate system, the longitude and latitude values ​​of each calibration coordinate in the spherical coordinate system are obtained. For two adjacent calibration coordinates, the longitude difference between them in the longitude direction is obtained. When obtaining the longitude difference, if the longitude values ​​of the two calibration coordinates cross the longitude boundary, the actual longitude difference after crossing the longitude boundary is taken as the longitude difference. The latitude difference between them in the latitude direction is also obtained. The longitude difference and latitude difference are compared with preset thresholds respectively. If the longitude difference exceeds the threshold, the calibration coordinate with the larger longitude value is marked as a rotation transformation singularity. If the latitude difference exceeds the threshold, the calibration coordinate with the larger latitude value is marked as a rotation transformation singularity. If both the longitude difference and latitude difference exceed the threshold, both calibration coordinates are marked as rotation transformation singularities.

[0007] As a further aspect of the present invention: constructing the new mesh patch includes: Local geometric features are extracted from the neighborhood of the rotation transformation singularity. An icosahedron is constructed with the rotation transformation singularity as the center. The direction of each vertex of the icosahedron is used as the initial direction template. The angle distribution of each ray in the initial direction template is redistributed according to the local geometric features to obtain an adaptive direction field. Twenty initial rays are obtained by drawing rays from the position of the rotation transformation singularity along each direction in the adaptive direction field. The intersection of each initial ray with the surface of the seabed basic model is used as the ray intersection point. Obtain the spatial straight-line distance between the ray intersection point and the location of the rotation transformation singularity. Obtain the average side length of the original mesh vertices adjacent to the rotation transformation singularity in the seabed base model at the location of the rotation transformation singularity. If the spatial straight-line distance exceeds three times the average side length, discard the ray intersection point. If the spatial straight-line distance does not exceed three times the average side length, retain the ray intersection point as a sampling point and use the sampling point as the base point for constructing the new mesh patch.

[0008] As a further aspect of the present invention, constructing the new mesh patch also includes: Obtain the position of the base point on the surface of the seabed base model, find the original mesh vertices adjacent to the base point. The original mesh vertices are the mesh vertices in the seabed base model that are closest to the base point in the direction of longitude or latitude along the surface of the seabed base model. Connect the base point with the adjacent original mesh vertices with straight line segments, and use the straight line segments as new mesh edges. The closed area formed by multiple new mesh edges and original mesh edges is used as a new mesh patch.

[0009] As a further aspect of the present invention: the modified marine platform main body model includes: Obtain the longitude and latitude values ​​of each vertex in the newly added mesh patch in the geographic spherical coordinate system. Convert the longitude and latitude values ​​into spatial rectangular coordinates along the Earth's radius to obtain spherical mapping coordinates. Obtain the inverse transformation matrix of the tilt attitude parameters. Multiply the spherical mapping coordinates with the inverse transformation matrix to obtain the vertex coordinates in the engineering rectangular coordinate system. Write the coordinates back to the vertex corresponding to the rotation transformation singularity in the main model of the offshore platform and update the patch index relationship with the vertex simultaneously to obtain the corrected main model of the offshore platform.

[0010] As a further aspect of the present invention: superimposing the modified marine platform main model with the seabed foundation model includes: The coordinate values ​​of all vertices in the corrected main model of the marine platform remain unchanged. The coordinates of the grid vertices in the seabed base model that have spatial overlap with the corrected main model of the marine platform are converted from the geographic spherical coordinate system to the engineering rectangular coordinate system. In the engineering rectangular coordinate system, the grid vertices of the seabed base model in the spatial overlap area are merged with the grid vertices of the corrected main model of the marine platform. During the merging, the coordinate values ​​of the grid vertices of the corrected main model of the marine platform are retained, and the grid vertices in the seabed base model that coincide with or are located inside the grid vertices of the corrected main model of the marine platform are deleted.

[0011] The beneficial effects of this invention compared to the prior art are as follows: This invention obtains the coordinates of tilted vertices by applying rotational and translational transformations of tilt attitude parameters to the initial vertex coordinates of the main model of the marine platform. These tilted vertex coordinates are then mapped to a geographic spherical coordinate system. During the mapping process, adjacent vertices with longitude or latitude differences exceeding a preset threshold are identified and marked as rotational transformation singularities. The main extension directions of the original mesh edges, the main local turning directions of the seabed foundation surface, and the main distortion directions of the singularity's neighborhood are extracted around the singularity. Using the vertices of a regular icosahedron as initial direction templates, the angle distribution of these templates is redistributed based on the aforementioned local geometric features, forming an adaptive direction field around the rotational transformation singularity. Rays are then emitted along each direction in this adaptive direction field, and new mesh patches are constructed through the intersections of these rays with the surface of the seabed foundation model. Finally, the local reconstruction results corresponding to these new mesh patches are inversely mapped back to the engineering rectangular coordinate system to replace the distorted mesh in the neighborhood of the rotational transformation singularity. Attached Figure Description

[0012] The invention will now be further described with reference to the accompanying drawings.

[0013] Figure 1 This is a flowchart illustrating a data-driven method for optimizing a three-dimensional model of a marine platform according to the present invention. Figure 2 This is a schematic diagram of the process for constructing new mesh patches according to the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Please see Figures 1-2 As shown, this invention is a data-driven method for optimizing a 3D model of a marine platform, comprising the following steps: Obtain the main model of the marine platform and the basic model of the seabed. The main model of the marine platform is constructed using an engineering rectangular coordinate system, while the basic model of the seabed is constructed using a geographic spherical coordinate system. Specifically, when acquiring the main model of the offshore platform, the vertex table, edge table, and patch table are read from the platform's 3D modeling file. The three coordinate values ​​of each vertex in the file are used as the initial vertex coordinates in the engineering Cartesian coordinate system. The engineering Cartesian coordinate system takes the platform's design reference point as the origin, the deck length direction as the X-axis, the deck width direction as the Y-axis, and the vertical upward direction as the Z-axis. The mesh connection relationship of the main platform is restored according to the patch index, ensuring that the order of adjacent vertices matches the storage order in the modeling file, facilitating subsequent point-by-point checking of coordinate jumps. When acquiring the seabed foundation model, longitude, latitude, and depth data are extracted from seabed survey data or seabed topographic mesh files. Longitude and latitude are used as horizontal positioning quantities in the geographic spherical coordinate system, and depth is converted into a radial position quantity relative to sea level. Then, mesh vertices and mesh patches of the seabed foundation model are generated according to the measurement point number or triangulation results. After reading, missing vertex checks, duplicate vertex merging, and illegal patch removal are performed on both the main platform model and the seabed foundation model. The corresponding number of each vertex, the numbers of adjacent vertices, and the number of the patch to which it belongs are saved.

[0016] The vertex coordinates of the main model of the offshore platform are corrected to obtain corrected coordinates, and then the corrected coordinates are mapped to the geographic spherical coordinate system; In a preferred embodiment of the present invention, obtaining the correction coordinates includes: Extract the vertex number, vertex coordinates, and patch index from the file. The vertex coordinates correspond to the geometric position of the main model of the offshore platform when it is in an upright position. The upright position is the state in which the deck plane coincides with the local horizontal plane and the axis of the pile legs extends vertically along the engineering rectangular coordinate system. Write the X, Y, and Z coordinates of each vertex into the vertex cache area, and establish a one-to-one correspondence between the vertex number and the coordinate value according to the storage order in the modeling file. Use this coordinate as the initial vertex coordinate.

[0017] When obtaining tilt attitude parameters, the rotation angles of the platform around the horizontal axis and the vertical axis of the engineering rectangular coordinate system are read from the installation condition data file, attitude measurement record or simulation input parameters. The horizontal axis adopts the deck length direction axis of the main model of the marine platform in the upright attitude, and the vertical axis adopts the vertical direction axis in the upright attitude. In order to avoid the vertex order disorder in the subsequent mapping stage, all vertices share the same set of tilt attitude parameters, and the angle values ​​are uniformly converted to the same angle system before storage.

[0018] Attitude transformation data is generated based on the two read rotation angles. During generation, the order of rotation around the horizontal axis first, then around the vertical axis, determines the arrangement of cosine and sine values ​​in each direction, and writes them into a 3x3 rotation matrix. Matrix multiplication is performed on the coordinates of each initial vertex to obtain the coordinates of intermediate vertices consistent with the current tilt attitude. After rotation, the translation parameters of the main model of the offshore platform in the engineering rectangular coordinate system are read. These translation parameters are the displacements of the platform reference point relative to the origin of the upright modeling in the X, Y, and Z directions. The coordinates of each intermediate vertex are added to the displacement in the corresponding direction to obtain the coordinates of the tilted vertex. To ensure that subsequent spherical coordinate mapping can track point by point, the original vertex number, original patch index, and transformed coordinate values ​​are retained for each tilted vertex coordinate generated, and this tilted vertex coordinate is determined as the correction coordinate.

[0019] In a specific embodiment of the present invention, when mapping the calibration coordinates to the geographic spherical coordinate system, the reference correspondence between the engineering rectangular coordinate system and the geographic spherical coordinate system is first determined. The reference correspondence uses a reference geographic point at the installation location of the offshore platform as the mapping origin. The reference geographic point takes the longitude, latitude, and local sea level elevation values ​​corresponding to the platform design center in the seabed foundation model, and the position of the reference geographic point in the engineering rectangular coordinate system is determined as the platform reference point.

[0020] After reading each calibration coordinate, the position increment of that calibration coordinate relative to the platform reference point is first calculated in three directions: east, north, and vertical. In the engineering rectangular coordinate system, the X direction corresponds to the local east, the Y direction to the local north, and the Z direction to the vertical elevation direction. Then, based on the longitude and latitude values ​​of the reference geographic point, the east position increment is converted into a longitude increment, the north position increment into a latitude increment, and the vertical position increment is superimposed with the elevation value of the reference geographic point to obtain the radial position quantity. During the conversion process, the direction unification processing is performed only once for the same vertex, ensuring that all vertices of the platform main model use the same positive longitude and positive latitude directions to avoid additional jumps due to inconsistent direction definitions between adjacent vertices. After obtaining the longitude and latitude increments, they are added to the longitude and latitude values ​​of the reference geographic point respectively to form the target longitude and target latitude values ​​corresponding to the calibration coordinate. Boundary normalization processing is performed on the target longitude value to ensure that the result falls within the preset longitude range, and polar restriction processing is performed on the target latitude value to keep the result within the legal latitude range.

[0021] After completing the single-point mapping, the obtained longitude, latitude, and radius position values ​​are written into the spherical coordinate cache according to the original vertex number, and the original patch index and adjacent vertex numbers corresponding to the vertex are retained. This allows for direct reading of adjacent grid connection pairs when checking the longitude and latitude differences of adjacent correction coordinates. To ensure that the platform main model and the seabed base model can be superimposed in the same geographic space, the longitude units, latitude units, and elevation datum used in the mapping are consistent with those of the seabed base model. When the seabed base model uses depth values, the depth values ​​are first converted into vertical position values ​​relative to the sea level before being stored in correspondence with the correction coordinate mapping results. This yields geographic spherical coordinate data that can be used for subsequent singularity identification and new grid patch construction.

[0022] Understandably, the main model of the offshore platform uses an engineering rectangular coordinate system during the modeling phase, which can easily express the relative positional relationships between the deck, legs, and components. However, the seabed foundation model is dependent on geographic space and can only correspond to the actual sea area location when placed in a geographic spherical coordinate system. Therefore, obtaining the calibration coordinates first and then performing the mapping essentially transforms the platform from a local geometric description under the design attitude to a real spatial description under the actual installation attitude, and then unifies this real spatial description with the geographic reference used by the seabed foundation. The rotation and translation in the calibration process do not change the shape of the platform itself, but rather move the vertices in the upright state to their actual positions after tilting and the legs are driven into the mud, so that each vertex carries working condition information. The mapping process is not a remodeling, but rather converts the vertices already in the actual attitude into latitude, longitude, and elevation expressions consistent with the seabed foundation, allowing data originally belonging to two different coordinate systems to be compared under the same spatial semantics. Only under this unified reference can the abnormal jumps of adjacent vertices in spherical coordinates be revealed. This is because implicit geometric distortion is not essentially an error in the coordinates of a single vertex, but rather a disruption of the local continuity of vertices after coordinate system rotation and spherical transformation. This is often difficult to detect when viewed solely within the engineering Cartesian coordinate system. After this processing, the correspondence between the platform main model and the seabed foundation model is clearly established. This allows for the accurate identification of rotational transformation singularities, the construction of transition meshes around these singularities, and the writing of the correction results back into the platform model. Thus, the original splicing relationship, which was merely surface alignment at the boundary, is transformed into a connection relationship where both geometric continuity and spatial positional relationship are simultaneously established.

[0023] Check whether the corrected coordinates jump during the mapping process, and mark the corrected coordinates that jump as rotation transformation singularities; In a preferred embodiment of the present invention, checking for the presence of a jump includes: Read the longitude and latitude values ​​corresponding to each vertex number from the spherical coordinate buffer, and then determine the two adjacent correction coordinates based on the edge table or patch index in the modeling file. The adjacent relationship is taken from the two vertices at both ends of the same mesh edge. If the modeling file does not store the edge table separately, the adjacent relationship is restored from the same patch according to the vertex connection order, and the same pair of adjacent vertices is checked only once.

[0024] After reading two adjacent calibration coordinates, first obtain the absolute difference between their longitude values. When the difference does not cross the longitude boundary, directly use the absolute difference as the longitude difference. When the difference crosses the longitude boundary, use the remainder after subtracting the absolute difference from the complete longitude circumferential angle as the actual longitude difference. This eliminates the apparent large span misjudgment caused by the longitude value changing from 179 to -179. Then obtain the absolute difference between their latitude values ​​as the latitude difference.

[0025] The preset thresholds use angle threshold data written to the parameter file before mapping checks. The corresponding thresholds are called for both longitude and latitude directions. During comparison, the actual longitude difference of each pair of adjacent vertices is compared with the longitude threshold, and the latitude difference is compared with the latitude threshold. If the actual longitude difference exceeds the longitude threshold, the corrected coordinate of the pair of adjacent vertices with the larger longitude value is written to the singularity marker table. If the latitude difference exceeds the latitude threshold, the corrected coordinate of the pair of adjacent vertices with the larger latitude value is written to the singularity marker table. If both the longitude and latitude differences exceed their respective thresholds, both pairs of adjacent vertices are written to the singularity marker table. When writing to the singularity marker table, the vertex number, original patch number, adjacent vertex numbers, corresponding longitude difference, and latitude difference are retained. Duplicate vertex numbers are deduplicated, ensuring that even if the same corrected coordinate is repeatedly identified in multiple adjacent relationships, only one rotation transformation singularity record is retained for direct use when constructing radial rays and adding new mesh patches centered on this singularity.

[0026] It is important to note that rotational transformation singularities are not arbitrary anomalies, but rather locations where the originally continuous vertex adjacency relationships of the platform's main model in the engineering Cartesian coordinate system exhibit local continuity distortion after tilting and mapping to the geographic spherical coordinate system. Adjacent vertices in the original model are connected by the same mesh edge, and the spatial distance and connection order should maintain a smooth transition. If, after mapping, the longitude or latitude values ​​suddenly jump beyond the normal range, it indicates that this is no longer a normal geometric unfolding result, but rather a local distortion, flip, or cross-boundary misalignment caused by the rotational transformation superimposed on the spherical coordinate representation. Using the difference in longitude and latitude between adjacent vertices as a criterion essentially checks whether changes in local mesh parameters still satisfy continuous geometric relationships. When crossing longitude boundaries, using the actual angular distance for judgment can distinguish between surface abrupt changes caused by coordinate representation and true geometric anomalies. After marking the jump point as the rotation transformation singularity, the newly added mesh patches can be corrected around the actual location of the distortion, rather than blindly reconstructing the entire platform. This way, implicit connection anomalies can be accurately extracted from the overall model, and the platform body and the seabed foundation can form a continuous geometric connection that can be used for subsequent simulation analysis under a unified coordinate reference.

[0027] Multiple rays extend outward from the rotational transformation singularity as the center, and new mesh patches are constructed based on the rays; In another preferred embodiment of the present invention, constructing the new mesh patch includes: First, using the projection point of the rotation transformation singularity onto the surface of the seabed base model as the center, find the target triangular facet containing this projection point. Then, include all adjacent triangular facets sharing edges with this target triangular facet in the local analysis region. Next, establish a local direction reference system, taking the projection point as the origin, and taking the east direction in the tangent plane of this point as the first direction, the north direction as the second direction, and the vertical direction as the third direction. Each original mesh vertex in the local analysis region is transformed to this local direction reference system, obtaining east, north, and vertical coordinates. All subsequent directions are only counted in the east and north planes, and all angles are converted to the undirected angle range of 0 to 180 degrees. Two vectors with opposite directions on the same extended axis are considered to be the same direction.

[0028] Subsequently, local geometric features are extracted within the local analysis region, including: The main extension direction of the original mesh edges is obtained as follows: First, read all existing mesh edges within the local analysis region. Existing mesh edges are limited to those whose two endpoints both belong to the local analysis region and are directly connected in the original mesh topology table; arbitrary temporary connections between two vertices are not allowed. For each existing mesh edge, read the east and north coordinates of the two endpoints in the local direction reference frame. Subtract the starting point from the endpoint to obtain the two-dimensional edge vector in the local plane. Then calculate the planar length and direction angle of this edge vector. Divide all existing mesh edges into eighteen direction buckets in ten-degree intervals. Each edge enters only one direction bucket; if the direction angle is on the bucket boundary, it is merged into the bucket with the smaller central angle. After bucketing, calculate a total weight for each direction bucket. The total weight is equal to the sum of the planar lengths of all edges within that bucket. Select the direction bucket with the largest total weight as the candidate bucket for the main direction. If two or more directional buckets have the same total weight, the bucket with the larger maximum plane length of a single edge within each bucket is prioritized. If they are still the same, the bucket with the larger number of edges is prioritized. If they are still the same, the bucket with the smaller central angle is selected. After determining the candidate buckets for the main direction, a unique representative direction is determined within that bucket. Specifically, the direction of each edge within the bucket is unified to be near the central angle of the candidate bucket. For example, if the central angle of the candidate bucket is 40 degrees, and the direction angle of an edge is 222 degrees, it is unified to 42 degrees. After unification, the unit direction vectors of all edges within the bucket are weighted and averaged according to their respective plane lengths to obtain an average direction vector. The angle corresponding to this average direction vector is then recorded as the main extension direction of the original mesh edge.

[0029] The main direction of local turning points on the seabed base surface is obtained as follows: First, read the coordinates of the three vertices of all triangular facets within the local analysis region, calculate the normal vector for each facet, and unify the direction of the normal vector to be outward. Then, find all shared edges within the local analysis region in the original mesh topology table; each shared edge corresponds to a pair of adjacent triangular facets. For each pair of adjacent triangular facets, calculate the angle between the two normal vectors; this angle is used as the turning strength value of the shared edge. Next, read the east and north coordinates of the two endpoints of the shared edge in the local direction reference frame to obtain the two-dimensional edge vector of the shared edge in the local plane. Since the turning point occurs in the direction crossing the shared edge, the direction of the shared edge itself is not taken; instead, the normal direction of the two-dimensional edge vector in the local plane is taken as the turning direction. Then, convert this turning direction into an undirected angle from 0 degrees to 180 degrees, and divide it into eighteen direction buckets in ten-degree intervals. Each shared edge enters a direction bucket, and the turning strength value of the shared edge is multiplied by the plane length of the shared edge as the contribution value of the shared edge to its respective direction bucket. For each directional bucket, the total contribution values ​​are summed to obtain the total turning weight of that bucket. The directional bucket with the largest total turning weight is selected as the candidate bucket for the main turning direction. If there is a tie, the maximum contribution value of a single shared edge, the number of shared edges, and the size of the bucket's central angle are compared sequentially within the candidate bucket, following the same rules as for the main extension direction. After selecting the candidate bucket, all turning directions within that bucket are unified to the vicinity of the candidate bucket's central angle. The unit vector of each turning direction is then weighted and averaged according to its respective contribution value to obtain an average direction vector. The angle corresponding to this average direction vector is then recorded as the local main turning direction of the seabed foundation surface.

[0030] The dominant distortion direction in the neighborhood of the rotation transformation singularity is determined by the directional distribution of the anomalous connecting edges between the singularity and its directly adjacent corrected coordinates. Based on the original mesh adjacency relationship, only adjacent corrected coordinates directly connected to the rotation transformation singularity via an original mesh edge are selected as the analysis objects. For each connecting edge pointing from the singularity to the adjacent corrected coordinate, the mapped longitude and latitude values ​​at both ends of the connecting edge are read, and the actual longitude difference and actual latitude difference are calculated according to the longitude boundary correction rule. The actual longitude difference is compared with the longitude threshold, and the actual latitude difference is compared with the latitude threshold. If neither exceeds the threshold, the connecting edge is not included in the distortion direction statistics; if at least one direction exceeds the threshold, the connecting edge is determined to be an anomalous edge. For each anomalous edge, the east and north coordinates of the singularity and adjacent corrected coordinates in the local direction reference frame are read, and the direction vector and direction angle of the anomalous edge in the local plane are calculated. The distortion intensity value is obtained through over-limit calculation. Specifically, longitude and latitude over-limit values ​​are calculated separately. Longitude over-limit is the positive value obtained by subtracting a longitude threshold from the actual longitude difference, and latitude over-limit is the positive value obtained by subtracting a latitude threshold from the actual latitude difference. If no limit is exceeded, the corresponding over-limit value is recorded as zero. Then, the longitude and latitude over-limit values ​​are transformed to the same local planar scale to form eastward and northward over-limit values. The length of the over-limit vector formed by these two values ​​is used as the distortion intensity value of the anomalous edge. All anomalous edges are assigned to corresponding direction buckets according to their direction angles, and the product of the distortion intensity value of each anomalous edge and its planar length is used as the contribution value of the anomalous edge to its direction bucket. For each direction bucket, the contribution values ​​of all anomalous edges within the bucket are accumulated one by one to obtain the total distortion weight of the direction bucket. The direction bucket with the largest total distortion weight is selected as the candidate bucket for the main distortion direction. If there are ties, the maximum contribution value of a single anomalous edge within the candidate bucket, the number of anomalous edges, and the size of the bucket's central angle are compared sequentially. After determining the candidate buckets, the directions of all anomalous edges within the buckets are made to be in the same direction, and the unit direction vectors are weighted by the contribution value of each anomalous edge to obtain the average direction vector. The angle value corresponding to this average direction vector is determined as the dominant distortion direction of the singularity neighborhood of the rotation transformation.

[0031] After extracting the main extension direction, local turning direction, and distortion main direction, a regular icosahedron is constructed based on the local direction reference frame. The spatial directions pointed to by the 20 vertices of the icosahedron are read and numbered sequentially as initial directions 1 to 20. The initial angular position of each initial direction in the local direction reference frame is also recorded. Then, a weight determination is performed on each initial direction individually. Specifically, the angles between the initial direction and the main extension direction, the local turning direction, and the distortion main direction are calculated, and each of these three angles is compared with a preset proximity interval. The high proximity interval is within 15 degrees, and the medium proximity interval is greater than 15 degrees but not greater than 30 degrees. For each initial direction, the interval to which its angle with the main extension direction, local turning direction, and distortion main direction belongs is determined. If the number of feature directions satisfying the high proximity condition is not less than 2, the initial direction is recorded as a high-weight direction; if the number of feature directions satisfying the high proximity condition is 1, the initial direction is recorded as a medium-weight direction; if the number of feature directions satisfying the high proximity condition is 0, and the number of feature directions satisfying the medium proximity condition is not less than 2, the initial direction is recorded as a medium-weight direction; if the number of feature directions satisfying the high proximity condition is 0, and the number of feature directions satisfying the medium proximity condition is less than 2, the initial direction is recorded as a low-weight direction.

[0032] After independently judging each of the 20 initial directions in the above manner, the weight level corresponding to each of the 20 initial directions is obtained. After the weight judgment is completed, while keeping the total number of directions unchanged at 20, the angular distribution range around each initial direction is redistributed. During the redistribution, the adjacency relationship is not determined according to the numbering order, but according to the spatial adjacency relationship of the vertices of the regular icosahedron in the local direction reference frame. The spatial adjacency relationship is the direction relationship between two vertices directly connected by the same edge in the regular icosahedron. Then, for each pair of spatially adjacent initial directions, their angular intervals are checked. If both pairs of spatially adjacent initial directions are high-weighted directions, the angular interval is reduced to make the direction distribution in the region denser; if both pairs of spatially adjacent initial directions are low-weighted directions, the angular interval is increased to make the direction distribution in the region sparser; if one of the pairs of spatially adjacent initial directions is a high-weighted direction and the other is a medium-weighted direction, the angular interval is slightly reduced; if one is a high-weighted direction and the other is a low-weighted direction, or one is a medium-weighted direction and the other is a low-weighted direction, the original angular interval is maintained or only slightly adjusted. After all adjacent angular intervals have been adjusted, the 20 adjusted directions are checked one by one to see if there are any direction intersections, direction inversions, or excessively small spacing. If the spacing between two directions is less than the allowable lower limit, one of the directions is slightly adjusted along the side away from the adjacent direction until all directions are orderly distributed and do not overlap. The 20 adjusted direction sets formed after the above processing are used as an adaptive direction field around the rotation transformation singularity. Subsequently, rays are emitted along the 20 directions in the adaptive direction field to obtain the intersection with the surface of the seabed basic model.

[0033] After obtaining the effective intersection points of each ray with the surface of the seabed base model, the corresponding seabed base model patch number is read, and the patch number is used as the entry point to look up the original mesh vertices directly connected to that patch in the mesh topology table of the seabed base model. The original mesh vertices are those that already exist in the vertex table when the seabed base model was initially modeled, excluding the base points in the newly added mesh patches and the replacement points generated during subsequent corrections. In specific processing, the target patch containing the projection position of the rotation transformation singularity is read first, and then the adjacent patches that share edges with the target patch are read. All the original vertex numbers appearing in the target patch and adjacent patches are summarized into a candidate vertex set. Then, based on the surface proximity relationship between the candidate vertices and the location of the rotation transformation singularity, the original mesh vertices adjacent to the rotation transformation singularity are screened out. The surface proximity relationship is based on the original mesh edge connection relationship. Vertices that are directly connected to the singularity region through one original mesh edge or through the same target patch are considered as adjacent original mesh vertices. After identifying adjacent original mesh vertices, the connection relationships between these vertices in the original mesh are read. Only original mesh edges whose endpoints both belong to the set of adjacent original mesh vertices are statistically analyzed. The spatial straight line length between the two endpoints of each original mesh edge is calculated, and the average of all obtained edge lengths is used as the average edge length of the seabed base model at the location of the rotation transformation singularity. For example, if four adjacent original mesh vertices are found in the neighborhood of a singularity, and there are five original mesh edges between these four vertices, the coordinates of the two endpoints of these five edges are read, and the edge length is calculated for each edge. The average of the five edge lengths is then taken to obtain the average edge length used for subsequent filtering. Then, the spatial straight-line distance between each ray intersection point and the location of the rotation transformation singularity is calculated, and the spatial straight-line distance is compared with 3 times the average side length. If the spatial straight-line distance exceeds the comparison value, it is determined that the ray intersection point deviates too far from the singularity neighborhood. Using the intersection point to construct a new mesh patch would cross an excessively large original mesh range, so the ray intersection point is discarded. If the spatial straight-line distance does not exceed the comparison value, the ray intersection point is retained as a sampling point, and the sampling point is written into the base point set.

[0034] After obtaining the set of base points, the system continues to read the patch number and vertex number of the seabed base model where each base point is located to determine its specific position on the seabed base model surface. Then, it searches for adjacent existing mesh vertices around the base point. Instead of using the arbitrarily closest principle, it searches for the closest mesh vertices to the base point along both the meridian and parallel directions on the seabed base model surface. Specifically, it reads the longitude and latitude values ​​of existing mesh vertices near the base point, selects vertices with smaller longitude differences and adjacent latitude differences as candidate vertices along the meridian direction, and vertices with smaller latitude differences and adjacent longitude differences as candidate vertices along the parallel direction. From these candidate vertices, it selects the vertex with the shortest surface distance along the corresponding direction as the adjacent existing mesh vertex. After determining the adjacent vertices, it connects the base point to the adjacent vertex with a straight line segment, and registers this straight line segment as a new mesh edge. If the same base point has adjacent vertices in both the meridian and parallel directions, corresponding new mesh edges are generated for each.

[0035] To ensure that newly added mesh edges can enclose a stable closed region, all base points are sorted according to their circumferential position relative to the singularity. The circumferential position is determined by the orientation of the base points in the local east-west and local north-north planes. After sorting, it is checked whether the newly added mesh edges corresponding to adjacent base points form closed loops with the original mesh edges. When two or more newly added mesh edges connect end-to-end with the original mesh edges of the seabed base model and form a closed boundary without self-intersection, the region inside the closed boundary is determined as a newly added mesh patch. If a base point only forms an isolated edge and cannot close with the surrounding boundary, its base point attributes are retained but a new mesh patch is not immediately generated. It is only included in the closure judgment after the adjacent base points are filled in. After the above processing is completed, the vertex coordinates of the newly added mesh patch are composed of the coordinates of the base point and the coordinates of the original mesh vertex connected to it. The boundary of the newly added mesh patch includes both the newly added mesh edge and the original mesh edge of the seabed base model. This ensures that when the platform main model at the singularity of the rotation transformation is subsequently corrected using the newly added mesh patch, the added transition region is consistent with the surface position of the seabed base model and is continuous with the original mesh topology.

[0036] Understandably, the rotational transformation singularity corresponds to the location where the local continuity is disrupted after mapping. If the original mesh connection method is still used at such locations, the anomaly will remain in the interface area between the platform and the seabed foundation. Therefore, subsequent processing cannot simply replace coordinates; it is also necessary to reorganize the local geometry around the anomaly center. The icosahedron provides an initial sampling skeleton with a relatively uniform directional distribution. Using it as a starting point ensures that all spatial orientations around the singularity are covered, and local surface information is not missed due to missing directions. An adaptive direction field is introduced on this basis because the distortion in the singularity's neighborhood, the original mesh orientation, and the undulations of the seabed surface are usually not uniform. If a completely uniform ray distribution is maintained, sampling resources will be wasted on directions with gentle changes, while insufficient coverage of the main distortion directions and local turning directions that truly need repair. Therefore, the ray angles are redistributed according to the main extension direction, local turning direction, and main distortion direction, which essentially allows the sampling density to change with the local geometric complexity. The intersection points are then filtered using the average edge length of their neighborhoods to limit the mesh patching range to an area within the original mesh scale that can withstand near the singularity. This prevents distant intersection points from being drawn into the current patching area, resulting in distorted, large-span connections. The subsequent process of finding and connecting the nearest original mesh vertex along the meridian or parallel direction is not simply about finding the nearest point, but rather about embedding the new boundary into the existing topology as closely as possible to the surface orientation of the original seabed base mesh. This results in new mesh patches that are both derived from real surface sampling and can be continuously stitched together with the original mesh. The earlier determination of the rotation transformation singularity relies on the anomaly in the latitude and longitude difference between adjacent vertices, essentially locating the source point of unstable local geometric continuity. The subsequent resampling and patching around this source point replaces the unstable region with a continuous transition surface constrained by the seabed base surface under the same geographic reference. Therefore, the entire process can converge the originally implicit local twists, flips, and misalignments to a repairable local range.

[0037] The main body model of the marine platform is corrected based on the newly added mesh patches. The corrected main body model of the marine platform is then superimposed on the seabed base model to output the optimized three-dimensional model of the marine platform.

[0038] In a preferred embodiment of the present invention, the modified marine platform main body model includes: The process reads the longitude, latitude, and radial position values ​​of all vertices in the newly added grid patch. The vertices in the new grid patch include the base point formed by the sampling points and the original grid vertices connected to the base point. For each vertex, a reverse conversion is performed using the reference geographic point, positive longitude direction, positive latitudinal direction, and elevation datum used when mapping the correction coordinates to the geographic spherical coordinate system. This converts the longitude and latitude values ​​into spatial rectangular coordinates along the Earth's radius, obtaining the spherical mapped coordinates of that vertex under a unified geographic reference. Subsequently, the tilt attitude parameters and translation parameters used when generating the correction coordinates are read, and an inverse transformation matrix is ​​constructed in the reverse order of the aforementioned correction process. This restores the spherical mapped coordinates point by point to their representation in the engineering rectangular coordinate system. Each spherical mapped coordinate is then multiplied by the inverse transformation matrix, and the corresponding translation is subtracted to obtain the coordinates of each vertex of the newly added grid patch in the engineering rectangular coordinate system.

[0039] After completing the coordinate back-calculation, the singularity vertex number, adjacent vertex numbers, and corresponding face number are read from the rotation transformation singularity record table. The vertex connectivity corresponding to the singularity is then read from the newly added mesh face record. The back-calculated coordinates corresponding to the singularity position are written back to the vertex table of the main marine platform model, and the original distorted vertex coordinates are updated. For the remaining vertices connected to the singularity's neighborhood in the newly added mesh face, the vertex index order in the corresponding face is updated according to the enclosure relationship between the new mesh edges and the original mesh edges, replacing the original abnormal connectivity relationships within the singularity's neighborhood with the connectivity relationships corresponding to the newly added mesh face. If the same rotation transformation singularity corresponds to multiple newly added mesh face vertices, the corresponding coordinates are written back sequentially according to the face adjacency relationship, and the index relationship is updated face by face. The vertex coordinates and face indices of vertices not marked as rotation transformation singularities remain unchanged, thus obtaining the corrected main marine platform model.

[0040] Another preferred embodiment of the present invention includes superimposing the modified marine platform main model with the seabed foundation model, comprising: When overlaying the corrected main model of the marine platform with the seabed foundation model, all coordinate values ​​in the vertex table of the corrected main model are kept unchanged. The outer bound of the model is used as the search reference for the overlapping area. Candidate mesh vertices falling within the outer bound and its neighboring range are read from the seabed foundation model. For each candidate mesh vertex, its longitude, latitude, and radial position are read. A reverse conversion is performed according to the same reference geographic point, the same positive longitude direction, the same positive latitudinal direction, and the same elevation reference used when mapping coordinates to the geographic spherical coordinate system. First, the longitude and latitude increments are restored to local eastward and northward displacements, and then the radial position is restored to the vertical displacement in the engineering rectangular coordinate system. This yields the transformed coordinates of the seabed foundation model vertices in the engineering rectangular coordinate system. The transformation relationship is the inverse of the aforementioned mapping relationship, thus ensuring that the spatial position of the seabed foundation model before and after the transformation is consistent with the reference of the main platform model before and after correction. After coordinate transformation, the positional relationship between candidate vertices of the seabed foundation model and the vertices of the corrected marine platform main model is compared one by one in the engineering rectangular coordinate system. When the spatial distance between the seabed foundation model vertex and the platform main model vertex is less than the preset overlap tolerance, the seabed foundation model vertex is determined to be an overlap vertex. When the seabed foundation model vertex does not overlap with the platform main model vertex, it is further determined whether the seabed foundation model vertex is located inside the platform main model based on the closed body enclosed by the platform main model facets. If it is inside, the vertex is marked as a vertex to be deleted. If it is neither overlapped nor located inside, it is retained as a valid vertex. Subsequently, vertex merging is performed on the mesh in the spatially overlapping area. The retained valid vertices of the seabed foundation model and the corrected marine platform main model vertices are written into a unified vertex table. For vertices determined to be overlapped, only the coordinate values ​​of the corrected marine platform main model vertices are retained, and the vertex indices in the relevant facets of the seabed foundation model are modified simultaneously. The seabed foundation model vertices determined to be inside the platform main model and their invalid facets are deleted. Finally, a continuously connected superimposed model is obtained in the engineering rectangular coordinate system.

[0041] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A data-driven method for optimizing a 3D model of a marine platform, characterized in that, Includes the following steps: Obtain the main model of the marine platform and the basic model of the seabed. The main model of the marine platform is constructed using an engineering rectangular coordinate system, while the basic model of the seabed is constructed using a geographic spherical coordinate system. The vertex coordinates of the main model of the offshore platform are corrected to obtain corrected coordinates, and then the corrected coordinates are mapped to the geographic spherical coordinate system; Check whether the corrected coordinates jump during the mapping process, and mark the corrected coordinates that jump as rotation transformation singularities; Multiple rays extend outward from the rotational transformation singularity as the center, and new mesh patches are constructed based on the rays; The main body model of the marine platform is corrected based on the newly added mesh patches. The corrected main body model of the marine platform is then superimposed on the seabed base model to output the optimized three-dimensional model of the marine platform.

2. The data-driven method for optimizing a three-dimensional model of a marine platform according to claim 1, characterized in that, Obtaining the calibration coordinates includes: Read the modeling file of the main offshore platform model. The vertex coordinates stored in the modeling file correspond to the geometric data of the main offshore platform model in an upright position. In the upright position, the deck plane of the main offshore platform model is parallel to the horizontal plane, and the leg axis of the main offshore platform model is perpendicular to the horizontal plane. Use the vertex coordinates read from the modeling file as the initial vertex coordinates. Obtain the tilt attitude parameters of the main offshore platform model. The tilt attitude parameters include the rotation angles around the horizontal axis and the rotation angles around the vertical axis in the engineering rectangular coordinate system. Construct a rotation matrix based on the rotation angles. Multiply the initial vertex coordinates by the rotation matrix to obtain the intermediate vertex coordinates. Obtain the translation parameters of the main offshore platform model in the engineering rectangular coordinate system. Add the intermediate vertex coordinates to the translation parameters to obtain the tilt state vertex coordinates. Use the tilt state vertex coordinates as the correction coordinates.

3. The data-driven method for optimizing a three-dimensional model of a marine platform according to claim 1, characterized in that, Checking for jumps includes: After mapping the calibration coordinates to the geographic spherical coordinate system, the longitude and latitude values ​​of each calibration coordinate in the spherical coordinate system are obtained. For two adjacent calibration coordinates, the longitude difference between them in the longitude direction is obtained. When obtaining the longitude difference, if the longitude values ​​of the two calibration coordinates cross the longitude boundary, the actual longitude difference after crossing the longitude boundary is taken as the longitude difference. The latitude difference between them in the latitude direction is also obtained. The longitude difference and latitude difference are compared with preset thresholds respectively. If the longitude difference exceeds the threshold, the calibration coordinate with the larger longitude value is marked as a rotation transformation singularity. If the latitude difference exceeds the threshold, the calibration coordinate with the larger latitude value is marked as a rotation transformation singularity. If both the longitude difference and latitude difference exceed the threshold, both calibration coordinates are marked as rotation transformation singularities.

4. The data-driven method for optimizing a three-dimensional model of a marine platform according to claim 1, characterized in that, Creating new mesh patches includes: Local geometric features are extracted from the neighborhood of the rotation transformation singularity. An icosahedron is constructed with the rotation transformation singularity as the center. The direction of each vertex of the icosahedron is used as the initial direction template. The angle distribution of each ray in the initial direction template is redistributed according to the local geometric features to obtain an adaptive direction field. Twenty initial rays are obtained by drawing rays from the position of the rotation transformation singularity along each direction in the adaptive direction field. The intersection of each initial ray with the surface of the seabed basic model is used as the ray intersection point. Obtain the spatial straight-line distance between the ray intersection point and the location of the rotation transformation singularity. Obtain the average side length of the original mesh vertices adjacent to the rotation transformation singularity in the seabed base model at the location of the rotation transformation singularity. If the spatial straight-line distance exceeds three times the average side length, discard the ray intersection point. If the spatial straight-line distance does not exceed three times the average side length, retain the ray intersection point as a sampling point and use the sampling point as the base point for constructing the new mesh patch.

5. The data-driven method for optimizing a three-dimensional model of a marine platform according to claim 4, characterized in that, Creating new mesh patches also includes: Obtain the position of the base point on the surface of the seabed base model, find the original mesh vertices adjacent to the base point. The original mesh vertices are the mesh vertices in the seabed base model that are closest to the base point in the direction of longitude or latitude along the surface of the seabed base model. Connect the base point with the adjacent original mesh vertices with straight line segments, and use the straight line segments as new mesh edges. The closed area formed by multiple new mesh edges and original mesh edges is used as a new mesh patch.

6. The data-driven method for optimizing a three-dimensional model of a marine platform according to claim 2, characterized in that, The revised main model of the offshore platform includes: Obtain the longitude and latitude values ​​of each vertex in the newly added mesh patch in the geographic spherical coordinate system. Convert the longitude and latitude values ​​into spatial rectangular coordinates along the Earth's radius to obtain spherical mapping coordinates. Obtain the inverse transformation matrix of the tilt attitude parameters. Multiply the spherical mapping coordinates with the inverse transformation matrix to obtain the vertex coordinates in the engineering rectangular coordinate system. Write the coordinates back to the vertex corresponding to the rotation transformation singularity in the main model of the offshore platform and update the patch index relationship with the vertex simultaneously to obtain the corrected main model of the offshore platform.

7. The data-driven method for optimizing a three-dimensional model of a marine platform according to claim 1, characterized in that, The modified main model of the marine platform is superimposed on the seabed base model, including: The coordinate values ​​of all vertices in the corrected main model of the marine platform remain unchanged. The coordinates of the grid vertices in the seabed base model that have spatial overlap with the corrected main model of the marine platform are converted from the geographic spherical coordinate system to the engineering rectangular coordinate system. In the engineering rectangular coordinate system, the grid vertices of the seabed base model in the spatial overlap area are merged with the grid vertices of the corrected main model of the marine platform. During the merging, the coordinate values ​​of the grid vertices of the corrected main model of the marine platform are retained, and the grid vertices in the seabed base model that coincide with or are located inside the grid vertices of the corrected main model of the marine platform are deleted.

Images