Method for automatic generation of structured surface mesh based on topology filling

Abstract

The application discloses a structure surface grid automatic generation method based on topological filling, and relates to the technical field of aerospace engineering simulation and grid generation. The method comprises the following steps: intelligently dividing a three-dimensional CAD model into parts to obtain parts with semantic labels and geometric contours; sequentially performing ordered topological filling on each part according to a preset order, adjusting the number of intersection edge points of adjacent parts to ensure grid connection, and generating topological lines; projecting the topological lines onto the model surface for calibration; adaptively interpolating the projected topological lines based on model curvature and simulation conditions to generate point cloud data; generating an initial quadrilateral grid based on the point cloud; projecting the initial grid nodes onto the model surface again and optimizing the quality, and outputting the final grid. The application realizes end-to-end automatic generation from a CAD model to a high-quality surface grid, eliminates manual intervention, ensures the natural matching of grids between parts, and significantly improves the grid generation efficiency and quality.

Description

Technical Field

[0001] This invention relates to the field of computational geometry and mesh generation, and more specifically, to an automated method for generating structural surface meshes based on topology filling. Background Technology

[0002] In computational fluid dynamics (CFD) and structural finite element analysis in aerospace engineering, the quality of the surface mesh directly determines the simulation accuracy and convergence. Currently, the mainstream surface structure mesh generation process relies heavily on human expert experience and usually includes the following steps: (1) studying the design documents and 3D digital model, deconstructing the simulation requirements and locking the key areas; (2) importing the CAD geometric model, manually checking and repairing geometric defects and simplifying minor features; (3) manually splitting functional components and sub-regions based on experience and marking the interfaces between adjacent components; (4) manually setting differentiated mesh parameters (size, quality threshold, element type) for each region; (5) manually selecting the filling template, generating the initial surface mesh for each region and fine-tuning the distorted elements in real time; (6) manually verifying the interfaces between adjacent components, adjusting the point matching and node alignment to ensure mesh connection; (7) evaluating the mesh quality, targeting and correcting substandard elements, and repeating the evaluation-correction until the key areas meet the standards; (8) globally optimizing the mesh size consistency and manually verifying that there are no omissions; (9) exporting the mesh and importing it into the simulation software, manually checking the compatibility, and reworking and correcting if necessary.

[0003] The following are the main technical problems and their causes in this manually-driven process:

[0004] (1) Highly subjective, cumbersome and error-prone: The core steps of mesh generation (feature selection, region splitting, parameter setting, and connection adjustment) all rely on expert experience and judgment, lack quantitative standards, and there are significant differences in operation among different experts; manual defect checking and edge-by-edge adjustment are labor-intensive, and hidden defects and minor misalignments are easily overlooked. The root cause is the lack of a quantitative correlation mechanism between geometric features, simulation requirements, and mesh parameters, and the absence of a standardized automated operation process.

[0005] (2) Long iteration cycle and extremely low R&D efficiency: If the simulation results are not up to standard or the requirements change, the parameters need to be readjusted, the region split, the mesh refilled and the verification repeated. The entire rework process is serial and the trial and error cost is high. The complete generation cycle of complex aircraft models can take 3-7 days, and a single rework takes 1-3 days. The root cause of the problem is that the mesh generation decision lacks foresight and cannot generate a high-quality mesh that meets the requirements at one time, relying on post-correction.

[0006] (3) Knowledge is difficult to accumulate and reuse, and the quality is inconsistent: the implicit experience of experts (such as the setting of key area parameters and distortion correction techniques) is difficult to standardize and document, and cannot be effectively transferred to the design of new models or new teams; the quality of meshes generated by different projects and engineers is uneven, and best practices at the organizational level cannot be accumulated and passed on. This is because experience has not been transformed into reusable algorithm rules, and there is a lack of a systematic framework for knowledge accumulation and reuse.

[0007] (4) Insufficient adaptation to complex geometry and obvious technical bottlenecks: For complex curved surfaces or multi-component interlaced structures such as stealth fighters and hypersonic aircraft, the difficulty of manual disassembly and filling increases exponentially, making it difficult to generate high-quality meshes without distortion; features such as abrupt curvature changes in key areas can easily lead to substandard mesh quality, and manual adjustment is difficult to completely solve the problem. The root cause is that the manual processing capability is limited by geometric complexity, and there is a lack of automated adaptation algorithms to cope with complex geometry.

[0008] In recent years, AI technology has been explored in the field of mesh generation, but most of these efforts have focused on post-hoc optimization. There is still a lack of fully automated mesh generation solutions that can replace human experts and integrate geometric features with simulation requirements, which cannot fundamentally solve core problems such as reliance on experience and low efficiency. Summary of the Invention

[0009] This invention aims to solve the following major problems existing in the prior art: low automation of structural surface mesh generation process, relying on manual intervention; poor mesh connection between adjacent components, requiring manual adjustment; insufficient fit between topology filling results and actual model, resulting in large geometric deviations; lack of adaptability in interpolation and size control, failing to match model curvature and computational requirements; and insufficient mesh quality optimization, making it difficult to meet the requirements of high-precision simulation.

[0010] To achieve the above-mentioned objectives, this invention provides an automated method for generating structural surface meshes based on topology filling, the method comprising:

[0011] Step 1: Obtain the 3D CAD model to be processed, perform component identification and subdivision on the 3D CAD model, and obtain multiple independent components and their geometric contour data;

[0012] Step 2: According to the preset component adjacency order, perform ordered topology filling on each component sequentially, including: based on geometric contour data, abstract the geometric contour of the current component to be processed into a polygon, and obtain the number of first vertices of the filled components adjacent to the current component on the intersecting edge; adjust the number of second vertices of the current component on the intersecting edge based on the number of first vertices, so that the number of second vertices is the same as the number of first vertices; use a predefined topology filling template to recursively subdivide the adjusted polygon until all sub-regions of the current component are quadrilaterals, and generate the topology line of the current component;

[0013] Step 3: Project the generated topology lines onto the surface of the 3D CAD model, perform geometric fit calibration, and obtain the projected topology lines;

[0014] Step 4: Perform adaptive interpolation processing on the projected topology lines. Based on the surface curvature of the 3D CAD model and the preset simulation calculation conditions, dynamically interpolate to generate point cloud data on the projected topology lines.

[0015] Step 5: Generate an initial structural surface mesh based on the point cloud data corresponding to each quadrilateral sub-region;

[0016] Step 6: Project the nodes of the initial structural surface mesh onto the surface of the 3D CAD model, optimize the quality of the projected mesh, and generate the final structural surface mesh.

[0017] The technical solution is summarized as follows:

[0018] This invention provides an automated method for generating structural surface meshes based on topology filling, comprising:

[0019] Step 1: Intelligent Component Subdivision—Through a combination of image recognition and geometric analysis, the system automatically identifies and subdivides each component of the structural model, obtaining component geometric contour data. Step 2: Ordered Topology Filling—Each component is abstracted into a polygon according to a preset adjacent order. The number of points at the intersection of adjacent components is adjusted collaboratively. A predefined topology filling template is used for recursive subdivision to generate topology lines. Step 3: Topology Line Projection Calibration—The filled topology lines are projected onto the actual model surface to eliminate geometric deviations. Step 4: Adaptive Interpolation Processing—Based on the model curvature and simulation calculation conditions, the topology lines are dynamically interpolated to generate point cloud data. Step 5: Quadrilateral Mesh Generation—An initial structural surface mesh is generated for each quadrilateral sub-region based on the point cloud data. Step 6: Mesh Projection and Optimization—The initial mesh is subjected to secondary projection and quality optimization to ensure mesh accuracy and quality.

[0020] The core principle of this invention is to construct a fully automated framework for the entire process of component partitioning, ordered filling, projection calibration, adaptive interpolation, mesh filling, mesh projection, and mesh optimization.

[0021] Component subdivision mechanism: Image recognition is used to locate the boundaries of the main structural components, and geometric feature analysis is combined to break them down into smaller components, achieving automated and accurate component identification. Ordered collaborative filling mechanism: Components are filled sequentially according to their adjacent order, and the number of intersecting points is adjusted collaboratively to ensure mesh connectivity between components. Geometric fitting mechanism: Topological line projection and secondary mesh projection are used to eliminate geometric deviations between the topological fill and the actual model. Adaptive control mechanism: The interpolation point density is dynamically adjusted based on model curvature and computational conditions to achieve personalized optimization of mesh size. Quality assurance mechanism: Through multi-dimensional quality index evaluation and iterative optimization, key quality parameters such as mesh orthogonality and smoothness are ensured to meet engineering requirements.

[0022] Preferably, step 1 specifically includes:

[0023] Step 1.1: Obtain the 3D CAD model to be processed, wherein the 3D CAD model has a boundary representation structure;

[0024] Step 1.2: Based on the engineering requirements of mesh generation, perform engineering semantic annotation on the faces in the boundary representation structure of the 3D CAD model to obtain component semantic labels; extract the geometric parameters of each face from the boundary representation structure; generate a multi-view 2D rendering image for each annotated face; associate and store the multi-view 2D rendering image, the geometric parameters, and the component semantic labels to form training samples;

[0025] Step 1.3: Train a multimodal recognition model using the training samples. The multimodal recognition model is configured to fuse input image features and language semantic information to output semantic labels for components on the surfaces of a 3D CAD model.

[0026] Step 1.4: Input the 3D CAD model to be identified into the trained multimodal recognition model, identify the faces in the 3D CAD model, and obtain the preliminary face-part semantic label mapping relationship;

[0027] Step 1.5: Based on the preliminary face-component semantic label mapping relationship, for faces whose recognition confidence output by the multimodal recognition model is lower than a preset threshold or whose geometric area ratio is less than a set threshold, the geometric rule auxiliary module is activated to perform semantic completion to obtain the final component segmentation result.

[0028] Step 1.6: Based on the final component segmentation result, extract the key geometric feature lines corresponding to each component from the boundary representation structure of the 3D CAD model. The component geometric contour data includes the final component segmentation result and the key geometric feature lines.

[0029] The method employs a trimodal fusion (vision-geometry-language) intelligent component recognition and segmentation approach. Specifically, it includes: performing engineering semantic annotation on the surfaces of the CAD model to generate multi-view 2D rendered images; using a multimodal recognition model containing a visual encoder, a language understanding module, and a cross-modal attention fusion module for recognition; initiating a geometric rule-assisted module to complete low-confidence or small-area surfaces; and finally outputting the component segmentation results and key geometric feature lines, which are then used together as the component's geometric contour data. This achieves full automation of component recognition, reducing the time-consuming manual work of several weeks to minutes. The output not only includes component attribution but also the key feature lines necessary for mesh generation, achieving readiness upon recognition and eliminating manual conversion steps. The geometric rule completion mechanism improves the completeness of recognition for small-sized, weakly featured components, ensuring that mesh generation covers all necessary geometric details.

[0030] Preferably, after step 1 and before step 2, the method further includes:

[0031] Based on the component semantic tags and a preset rule base, the component generation order of the overall machine surface mesh is determined, resulting in one or more component generation sequences. After step 1 and before step 2, based on the component semantic tags obtained in step 1 and a preset rule base (such as aerodynamic criticality priority, geometric dependencies, etc.), the component generation order of the overall machine surface mesh is determined, resulting in one or more component generation sequences. This solves the problem of chaotic component mesh generation order and lack of global planning in traditional methods, leading to difficulties in post-processing. Intelligent planning of the generation order provides a logical basis for subsequent orderly filling and constraint transfer; prioritizing critical or complex components allows subsequent components to inherit the boundary constraints of already generated components, ensuring the overall coordination of the mesh at the process level and reducing local adjustments.

[0032] Preferably, step 4 includes:

[0033] The component to be generated is determined according to the component generation sequence. The topological structure block, semantic label and working condition parameters of the component to be generated are input into the trained graph neural network (GNN) model. The GNN model collaboratively predicts the size distribution function of all topological edges and outputs the initial node distribution of each topological edge. The topological structure block is composed of the topological line generated in step 2 and the key geometric feature line obtained in step 1.

[0034] For subsequent components in the component generation sequence, the node coordinate sequence on the shared topological edge is inherited from the generated adjacent component mesh as a rigid constraint. Based on the rigid constraint and the size distribution function predicted by the graph neural network (GNN) model, the node distribution of the unconstrained edge is determined. The interpolation point density is adjusted to generate point cloud data in combination with the surface curvature and preset simulation calculation conditions.

[0035] Step 4 is specifically defined as follows: Based on the component generation sequence, the topological structure block, semantic label, and operating condition parameters of the first component to be generated are input into a trained Graph Neural Network (GNN) model. The GNN predicts the size distribution function of all topological edges and outputs the initial node distribution. For subsequent components, the node coordinates on the shared topological edges are inherited from the meshes of adjacent generated components as rigid constraints. The node distribution of unconstrained edges is determined according to the size distribution function predicted by the GNN, and the interpolation density is adjusted in combination with curvature and operating conditions. This solves the problems of traditional methods, such as the reliance on manual experience to set the mesh size distribution, the inability to take into account both component semantics and operating condition requirements, and the need for repeated manual debugging of node matching between components. The GNN model integrates multi-source information such as geometry, semantics, and operating conditions, realizing intelligent collaborative prediction of size distribution. This ensures that the node distribution of the first component automatically meets the requirements of intra-block edge matching and inter-block connection. Subsequent components are generated under rigid constraints, ensuring natural matching of the entire machine mesh on the common boundary without manual debugging. The adaptive adjustment of curvature and operating conditions makes the mesh density distribution highly consistent with the physical field characteristics.

[0036] Preferably, in step 2, a predefined topology filling template is used to recursively subdivide the adjusted polygon, including:

[0037] The variable-length vertex coordinate sequence of the current polygon to be processed is normalized into a fixed-dimensional state vector through zero-padding and mask marking;

[0038] The fixed-dimensional state vector is input into the trained deep reinforcement learning model. The deep reinforcement learning model selects the current action to be executed from a predefined set of topological filling template actions that cover triangles to hexagons based on a preset vertex scheduling strategy. The current action is then applied to geometrically segment the polygon to be processed, generating several sub-polygons and identifying quadrilateral sub-regions within them.

[0039] Specifically, the recursive partitioning in step 2 is defined as follows: the variable-length vertex coordinate sequence is normalized to a fixed-dimensional state vector through zero-padding and masking; this vector is then input into a trained deep reinforcement learning model, which selects the current action from a predefined set of topology-filling template actions covering triangles to hexagons based on a vertex scheduling strategy; the action is applied for geometric segmentation, generating sub-polygons and identifying quadrilateral sub-regions within them. This addresses the problems of traditional topology-filling methods relying on fixed rules, difficulty adapting to varying geometric shapes, and unstable partitioning quality. The deep reinforcement learning model, trained on a large amount of data, learns to adaptively select the optimal partitioning template based on different polygon shapes, avoiding the limitations of manually preset rules; the masking mechanism solves the variable-length input problem, ensuring the model's applicability; and the final generated quadrilateral mesh significantly outperforms traditional heuristic methods in terms of orthogonality, smoothness, and other quality metrics.

[0040] Preferably, in step 2, the number of first vertices of the filled components adjacent to the current component on the intersecting edge is obtained; the number of second vertices of the current component on the intersecting edge is adjusted based on the number of first vertices, specifically as follows:

[0041] Read the vertex coordinate sequence of the intersection edge between the filled component and the current component, and count the number of vertices as the first vertex count; expand or reduce the vertex count of the intersection edge corresponding to the current component from the initial value to be equal to the first vertex count, and recalculate the adjusted coordinate position of each vertex based on linear interpolation or geometric feature resampling algorithm to realize the one-to-one correspondence between points on the intersection edge.

[0042] In step 2, the following steps are specified: The vertex coordinate sequence of the intersection edge between the filled component and the current component is read, and the number of vertices is counted as the first vertex count. The vertex count of the corresponding intersection edge of the current component is expanded or reduced from the initial value to be equal to the first vertex count, and the vertex coordinates are recalculated based on linear interpolation or geometric feature resampling. This solves the problem of mismatched point counts on the intersection edge of adjacent components in existing technologies, requiring manual adjustment edge by edge. It achieves automatic and accurate matching of point counts on the intersection edge, ensuring a one-to-one correspondence between mesh nodes on both sides, fundamentally eliminating mesh misalignment and gaps. The resampling algorithm ensures that the geometric position of the adjusted vertices matches the component contour, without affecting the internal mesh quality of the component.

[0043] Preferably, step 4 involves adaptive interpolation processing of the projected topology lines, specifically including:

[0044] The curvature distribution of the surface of the 3D CAD model corresponding to the projected topology line is calculated. The interpolation point density is increased in the region where the curvature is greater than the first preset threshold, and the interpolation point density is reduced in the region where the curvature is less than the second preset threshold. Based on the working parameters of the subsequent simulation calculation, the minimum size threshold and the maximum size threshold of the interpolation points are adjusted so that the generated mesh size meets the calculation accuracy requirements.

[0045] Specifically, step 4 involves calculating the curvature distribution of the 3D CAD model surface corresponding to the projected topology lines, increasing the interpolation point density in areas of high curvature, and decreasing the interpolation point density in areas of gentle curvature; simultaneously, adjusting the size threshold of the interpolation points based on the operating parameters of subsequent aircraft simulation calculations (such as turbulence model requirements and stress gradient requirements). This solves the problems of traditional interpolation methods, such as fixed mesh size and inability to adapt to changes in geometric curvature and simulation calculation requirements. The generated mesh is automatically refined in areas rich in geometric details, accurately capturing surface features; it is sparsely distributed in flat areas, controlling the total mesh size; and it simultaneously meets the resolution requirements of simulation calculations for key areas, achieving a unification of geometric and physical adaptation.

[0046] Preferably, step 6 involves quality optimization of the projected mesh, specifically including:

[0047] Calculate the orthogonality, smoothness, aspect ratio, and convexity quality indices of the grid cells, obtain the index calculation results, and evaluate whether each grid cell meets the preset quality threshold based on the index calculation results;

[0048] For mesh cells that do not meet the preset quality threshold, adjustments are made by moving node positions, refining local meshes, or merging them.

[0049] Repeat the evaluation and adjustment process until the quality indicators of all grid cells meet the preset quality threshold.

[0050] Step 6 clarifies the specific process of quality optimization: calculating the orthogonality, smoothness, aspect ratio, and convexity quality indicators of the mesh elements; evaluating whether each element meets the preset quality threshold based on these indicators; adjusting elements that do not meet the threshold by moving node positions, local mesh refinement, or merging operations; repeating the evaluation and adjustment until all elements meet the standards. This addresses the problems of disconnect between mesh quality evaluation and optimization, unclear iterative processes, and difficulty in guaranteeing the final mesh quality in existing technologies. By clearly separating quality evaluation and adjustment actions, a closed-loop iterative optimization process is formed, ensuring that all quality indicators of the final mesh meet engineering requirements; multi-dimensional indicators comprehensively reflect the geometric adaptability of the mesh, avoiding the limitations of single-indicator optimization.

[0051] Preferably, the geometric rule auxiliary module determines and corrects the semantic labels of the surface components based on the geometric topological relationships and parametric features of the 3D CAD model, thereby forming the final component segmentation result.

[0052] Preferably, the extraction of key geometric feature lines corresponding to each component includes:

[0053] For the components identified by the multimodal recognition model, by analyzing the curvature distribution and boundary topology of its constituent surfaces, a geometric algorithm is used to extract the key geometric feature lines of the components, including its boundary lines and feature contour lines.

[0054] For the feature surfaces identified by the geometric rule assistance module, the corresponding key geometric feature lines are extracted based on their geometric definitions and their relationship with adjacent surfaces.

[0055] Specifically, for the main structural components identified by AI, geometric algorithms such as curvature and topology analysis are used to automatically extract their boundaries and contours. For small features identified by rules, extraction is performed directly based on their geometric definitions (such as the intersection of two surfaces). Both methods ensure that the output feature lines are accurately associated with the semantic labels of the components. This achieves automated integration from semantic segmentation to geometric feature extraction, outputting geometric data directly relied upon for operations such as mesh encryption and block division.

[0056] Preferably, step 5, generating the initial structural surface mesh, specifically includes:

[0057] For each quadrilateral sub-region, based on the point cloud data on its boundary, uniformly distributed grid nodes are generated inside the sub-region using interpolation; the grid nodes are connected to form quadrilateral grid cells, and the vertex index and coordinate information of each grid cell are recorded.

[0058] Specifically, step S5 defines the following: For each quadrilateral sub-region, based on the point cloud data at its boundary, uniformly distributed mesh nodes are generated within the sub-region using interpolation; the nodes are connected to form quadrilateral mesh cells, and the vertex index and coordinate information of each cell are recorded. This solves the problems of uneven node distribution and cell distortion when generating meshes from point cloud data. The interpolation method ensures a continuous transition between internal and boundary nodes, resulting in uniformly sized and regularly shaped mesh cells; recording the vertex index provides a complete data structure for subsequent quality optimization and output.

[0059] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0060] Because this invention employs a fully automated technical solution encompassing intelligent component partitioning, ordered topology filling, line projection calibration, adaptive interpolation, and mesh optimization, it achieves the following significant technical effects:

[0061] Significantly improved automation: Through intelligent component segmentation and orderly collaborative filling based on image and geometry, the system eliminates the reliance on manual intervention, achieving full automation from component recognition to mesh generation, and greatly improving generation efficiency.

[0062] Component connectivity optimization: The point coordination mechanism between adjacent components ensures seamless mesh connection, avoiding the connection breakage problem in traditional methods and reducing the workload of subsequent adjustments.

[0063] Improved geometric fit accuracy: The two projection calibration steps (topology line projection + mesh projection) effectively eliminate the geometric deviation between topology filling and the actual model, ensuring accurate fit between the mesh and the structural surface.

[0064] Enhanced Adaptive Capability: Based on adaptive interpolation and size control of model curvature and computational conditions, the mesh can capture complex geometric details and match subsequent simulation requirements, achieving personalized optimization.

[0065] Stable and reliable mesh quality: A multi-dimensional quality assessment and optimization mechanism ensures that key indicators such as mesh orthogonality and smoothness meet the standards, providing a guarantee for high-precision simulation.

[0066] High generalizability: This process can be adapted to different types and complex structural models without significant parameter adjustments, and has good engineering applicability. Attached Figure Description

[0067] The accompanying drawings, which are provided to further illustrate embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention.

[0068] Figure 1 Flowchart of an automated method for generating structural surface meshes based on topology filling;

[0069] Figure 2 This is a schematic diagram of the component assembly;

[0070] Figure 3 A schematic diagram of a triangle topology filling template;

[0071] Figure 4 A schematic diagram of a quadrilateral topology-filling template;

[0072] Figure 5 A schematic diagram of a pentagonal topology filling template;

[0073] Figure 6 Schematic diagram of a hexagonal topology filling template;

[0074] Figure 7 This is a schematic diagram of the Q-network model structure;

[0075] Figure 8 This is a schematic diagram of the segmentation of the polygon to be processed. Detailed Implementation

[0076] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other.

[0077] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0078] Example 1;

[0079] Please refer to Figure 1 , Figure 1 The flowchart of the automated generation method for structural surface mesh based on topology filling is shown in Embodiment 1 of the present invention. The method flow is as follows:

[0080] This method achieves a complete process for automated generation of surface meshes for aircraft structures, including the following steps:

[0081] 1) Intelligent component subdivision: Based on a combination of image recognition and geometric analysis, the various components of an aircraft model are automatically subdivided;

[0082] 2) Ordered topology filling: Abstract each component into a polygon in a preset adjacent order, coordinately adjust the number of points at the intersection of adjacent components, and perform topology filling;

[0083] 3) Topology line projection calibration: Project the filled topology lines onto the actual aircraft model surface to ensure geometric fit;

[0084] 4) Adaptive interpolation processing: Based on the model curvature and calculation conditions, point interpolation and adaptive size control are performed on the topology lines;

[0085] 5) Quadrilateral mesh generation: Generate an initial mesh for each quadrilateral region using interpolation;

[0086] 6) Mesh projection and optimization: Perform secondary projection and quality optimization on the initial mesh to ensure mesh accuracy and quality.

[0087] Detailed explanation of each step:

[0088] Step 1: Intelligent component breakdown:

[0089] Execution entity: Component identification and segmentation module;

[0090] Execution content: Automatically identify components such as the nose, fuselage, wings, and tail of an aircraft model, and achieve precise dissection;

[0091] Specific implementation:

[0092] The main structural components refer to the components with independent functions on the aircraft, such as the entire wing, the entire fuselage, and the tail. For simplicity, they will be referred to as large components in the following description. Large component identification: Image recognition algorithms (such as convolutional neural networks) are used to analyze the overall model image of the aircraft, locate the outline boundaries of large components such as the nose, fuselage, wings, and tail, and output the preliminary division results of large components.

[0093] Small component decomposition: Based on geometric feature analysis (such as curvature abrupt change point detection and boundary line continuity analysis), large components are further decomposed, such as the wing being decomposed into small components such as wing root, wing body, and wingtip, to ensure that the component division meets the mesh generation requirements;

[0094] Component information storage: Records the geometric contour data of each component and the relationship between adjacent components, providing a foundation for subsequent orderly filling.

[0095] Step 2: Ordered Topology Fill; Input: Component geometric contour data and relationships between adjacent components output from Step 1; Preset component adjacency order (e.g., nose-fuselage-wing-tail, determined by subsequent Step 3, assumed here is already given). Output: Topology lines for each component (i.e., boundary lines of quadrilateral regions after internal subdivision of the component).

[0096] Execution subject: Topology filling module; Execution content: Process each component sequentially according to a preset order (such as nose-fuselage-wing-tail) to achieve orderly topology filling and connection between components;

[0097] Specific implementation:

[0098] Part polygon abstraction: The geometric contour of each subdivided part is abstracted into a polygon, and the vertex coordinates and quantity information are extracted;

[0099] Adjacent component point coordination: Components are processed sequentially according to a preset order. For the current component, its intersecting edges with all filled components are checked. The vertex coordinate sequence of the filled components on the intersecting edges is read, and the number of vertices is counted as the first vertex count. The vertex count of the current component's corresponding intersecting edge is expanded or reduced from the initial value to be equal to the first vertex count, and the adjusted coordinate positions of each vertex are recalculated based on linear interpolation or geometric feature resampling algorithms to ensure a one-to-one correspondence between points on the intersecting edges. For example... Figure 2 As shown, Figure 2 The diagram shows the component assembly. The nose and fuselage intersect. There are already 4 points on the nose side. The corresponding points on the fuselage side are expanded from the initial 2 points to 4 points. The positions of the newly added points are determined by interpolation.

[0100] Topology fill execution: A polygon subdivision method is used to recursively subdivide the adjusted polygons until all sub-regions are quadrilaterals. Nine topology fill templates are used, and the appropriate template is selected based on the number of vertices of the current component's polygons to complete the sub-region topology fill. This process is repeated until all sub-regions of the current component are quadrilaterals.

[0101] Sequential Iterative Processing: All components are processed sequentially according to a preset order, ensuring that each pair of adjacent components achieves point-to-point coordination and seamless connection. Finally, step 2 outputs the topology line of each component, which is the set of boundary lines of all quadrilateral sub-regions within the component.

[0102] Step 3: Topology line projection calibration:

[0103] Execution Module: Projection Calibration Module; Execution Content: Project the topology lines of all filled components onto the surface of the actual 3D model of the aircraft to eliminate geometric deviations; Input: Component topology lines output from step 2; Original 3D CAD model of the aircraft. Output: Projected topology lines.

[0104] Specific implementation:

[0105] Determining the projection datum: Establish a projection coordinate system based on the surface geometric data of the actual 3D model of the aircraft; the world coordinate system of the model is usually used.

[0106] Topology line projection: Using orthographic or oblique projection algorithms (oblique projection can also be selected based on curvature changes; the projection algorithm can be modified according to actual conditions, and this embodiment does not impose any limitations; a fast ray intersection algorithm based on AABB trees can be used to achieve efficient projection), each topology line generated by topology filling is projected onto the actual surface of the corresponding component, ensuring that the topology line perfectly matches the geometry of the model surface; the projection direction is generally taken as the normal to the plane containing the topology line, ensuring that the projection point falls on the model surface. For complex curved surfaces, the ray intersection method can be used to calculate the projection point.

[0107] Post-projection verification: Calculate the distance deviation between the projected topology lines and the model surface (e.g., the average or maximum of the shortest distances from each point to the model). If the deviation exceeds a preset threshold (e.g., 0.01mm), readjust the projection parameters (e.g., projection direction, number of iterations) and project again until the accuracy requirements are met. Finally, step 3 outputs the projected topology lines that precisely fit the model surface.

[0108] Step 4: Adaptive interpolation processing:

[0109] Execution Module: Interpolation and Size Control Module; Execution Content: Performs point interpolation on the projected topology lines and implements adaptive size control based on model curvature and calculation conditions; Input: Projected topology lines output from step 3; component semantic tags; simulation operating parameters (such as Mach number, Reynolds number, turbulence model requirements, etc.). Output: Point cloud data generated by interpolation on the topology lines (the node coordinate sequence of each topology edge).

[0110] Specific implementation:

[0111] Preliminary planning of interpolation points: Based on the constraint that the number of edge points of the structural mesh is equal, the number of interpolation points on each topological line is initially determined;

[0112] Adaptive curvature adjustment: Calculates the curvature distribution of the model surface corresponding to the projected topology lines. For regions with curvature greater than a first preset threshold (e.g., 0.05) (e.g., wing leading edge, nose tip), the interpolation point density is increased; for gentler regions with curvature less than a second preset threshold (e.g., 0.01) (e.g., fuselage midsection), the interpolation point density is decreased. Density adjustment can be achieved through local subdivision or merging to ensure the mesh accurately captures geometric details. Curvature can be calculated using Gaussian curvature or average curvature, estimated from discrete point clouds of the model surface, or directly using the curvature query function provided by the CAD kernel. The curvature threshold can be set empirically, for example, a first threshold of 0.05 and a second threshold of 0.01, but needs to be adjusted according to the specific model.

[0113] Computational condition adaptation: Based on simulation requirements (e.g., turbulence models require wall mesh y+ values, structural analysis requires stress gradient resolution), adjust the minimum and maximum size thresholds of the interpolation points. For example, for high Reynolds number flows, the leading edge of the wing needs to be fined to the order of 0.1 mm; for regions far from the wall, the size can be relaxed to 10 mm. Size constraints limit the spacing between interpolation points, ensuring that the generated mesh size satisfies a balance between computational accuracy and efficiency.

[0114] Interpolation execution: Using cubic spline interpolation or linear interpolation algorithms, a preset number of points are inserted into each topological line to generate complete topological point cloud data. The interpolation process must ensure that the number of nodes on opposite edges is equal (required for structured meshes). Finally, step 4 outputs the node coordinate sequence of all topological edges for each component.

[0115] The initial planning of interpolation points can be achieved using a GNN to predict the initial node distribution: the component topology blocks are transformed into graph structure data, with topological corners as nodes and topological edges as edges. The component semantic labels and operating parameters are embedded as features of the nodes and edges, and then input into a trained graph neural network (GNN) model. The GNN collaboratively predicts the size distribution function of all topological edges and outputs the initial node distribution (including the number of nodes and their approximate positions) for each topological edge.

[0116] The specific implementation method is as follows:

[0117] 1. Graph Structure Transformation of Topological Blocks: The component topological blocks generated in step 2 are transformed into graph structure data, which is then used as input to the GNN model. The specific transformation method is as follows:

[0118] Node Definition: Each corner point in the topology block is defined as a node in the graph. A corner point is a vertex where topological edges intersect, i.e., a vertex of a quadrilateral logical unit. Edge Definition: Each topological edge connecting two corner points is defined as an edge in the graph. The direction of the edges can be set to undirected or directed as needed (undirected graphs are usually used because mesh topologies have symmetry). Graph Topology: The connections between quadrilateral logical units in the topology block are preserved, i.e., corner points connected by shared edges are indirectly connected in the graph, forming a complete graph structure.

[0119] 2. Node Feature Embedding: Construct a feature vector for each node (corner point), containing the following information: Corner geometric features: the corner point's coordinates (x, y, z) in 3D space; the curvature at the corner point (which can be calculated from the Gaussian curvature or average curvature of the CAD model surface); the corner point type (e.g., boundary corner, interior corner, feature line intersection, etc.). Component semantic labels: The component semantic labels obtained in step 1 (e.g., wing leading edge, fuselage midsection, etc.) are one-hot encoded or embedded and used as the node's category features. Adjacency information: the node's degree (the number of connected topological edges), reflecting its importance in the topological structure.

[0120] 3. Edge Feature Embedding: Construct a feature vector for each edge (topological edge), containing the following information: Edge geometric features: edge length; edge curvature distribution (curvature variation along the edge); edge type (e.g., straight edge, curved edge, feature line, etc.). Component semantic labels: semantic labels of the component to which the edge belongs (shared with nodes). Operating condition parameters: Encode user-input simulation operating condition parameters (e.g., Mach number, Reynolds number, turbulence model type, target y+ value, etc.) into edge features, enabling the GNN to adjust size predictions according to operating conditions. Adjacency relationship features: the types of the two nodes connected by the edge and the angular relationship between them.

[0121] 4. GNN Model Architecture Design: A GNN architecture suitable for graph regression tasks is adopted, with the following specific design: Input Layer: Receives node feature matrices (dimension: number of nodes × node feature dimension) and edge feature matrices (dimension: number of edges × edge feature dimension). Graph Convolutional Layers: Stacked 3-5 graph convolutional layers to aggregate neighborhood information. Each layer can use operators such as GraphSAGE, GAT (Graph Attention Network), or GIN (Graph Isomorphism Network).

[0122] Edge feature fusion: In each graph convolutional layer, edge features are fused with node features. This can be done by concatenation or addition, allowing the model to utilize information from both nodes and edges simultaneously.

[0123] Output layer: Design two parallel output heads:

[0124] Size distribution prediction head: Outputs a size distribution function for each edge, usually represented as the parameters (mean, variance) of a Gaussian mixture model or a discretized size probability distribution.

[0125] Node distribution prediction head: Outputs the initial node distribution for each topological edge for the first component, including the number of nodes (integer) and the approximate location of the nodes (parametric coordinates along the edge).

[0126] 5. Loss function design:

[0127] When training the GNN model, a multi-task loss function is used:

[0128] ;

[0129] in, For size distribution prediction loss, negative log-likelihood loss (for probability distribution output) or mean squared error loss (for deterministic output) can be used. The node distribution prediction loss includes the cross-entropy loss for predicting the number of nodes and the mean squared error loss for predicting the node location. This is a regularization term to prevent overfitting. , , These are weighting coefficients, which can be adjusted based on the validation set; they are typically set to [value missing]. =1.0, =0.5, =0.01.

[0130] Training Data Construction: Training data is collected from historical mesh generation projects, including a large number of topological blocks of aircraft components, corresponding final mesh data, and operating parameters at the time of generation. The actual node distribution (number and location of nodes) on each topological edge is extracted from the final mesh as supervision labels. The topological blocks are transformed into a graph structure, and node and edge features are constructed to form training samples. Data Augmentation: Minor geometric deformations are performed on the topological blocks to generate more diverse training samples.

[0131] Prediction and Output: In the application phase, for a given component: its topological structure blocks are transformed into a graph structure as described above, constructing node features and edge features. The trained GNN model is input and forward propagated. The size distribution function and initial node distribution of each topological edge are obtained from the output layer. For the first component, the predicted initial node distribution is directly used as the basis for subsequent interpolation. For subsequent components, the size distribution function predicted by the GNN is combined with the rigid constraints inherited from adjacent components to determine the node distribution of unconstrained edges.

[0132] Step 5: Generating the quadrilateral mesh:

[0133] Execution Module: Mesh Generation Module; Execution Content: Based on the interpolated topological point cloud, generate an initial structural surface mesh for each quadrilateral sub-region; Input: Point cloud data output from step 4; Component topology. Output: Initial structural surface mesh (quadrilateral mesh).

[0134] Specific implementation:

[0135] Region division confirmation: Determine the boundary point cloud coordinates of each quadrilateral sub-region. Typically, each sub-region is enclosed by four topological edges, and the coordinates of the nodes on each edge are already determined.

[0136] Mesh node generation: Uniformly distributed mesh nodes are generated within the quadrilateral sub-region using bilinear interpolation or Coons surface interpolation. For example, if the number of nodes on opposite sides is m and n respectively, the number of internal nodes can be set to (m-1)×(n-1). The coordinates of each node are calculated using the interpolation formula to ensure continuity with the boundary nodes.

[0137] Initial mesh construction: Connect mesh nodes to form quadrilateral mesh cells, and record the vertex index and coordinate information of each cell. At the same time, establish the adjacency relationship between cells and edges to provide a data foundation for subsequent optimization.

[0138] Step 6: Mesh Projection and Optimization

[0139] Execution Module: Mesh Optimization Module; Execution Content: Perform secondary projection and quality optimization on the initial mesh to ensure mesh quality and accuracy; Input: Initial mesh output from step 5; Original 3D CAD model. Output: Final structural surface mesh.

[0140] Specific implementation:

[0141] Secondary projection calibration: All nodes of the initial mesh are projected again onto the surface of the actual aircraft model to further eliminate the deviation between the mesh and the geometric model. The projection method is the same as in step 3.

[0142] Mesh quality assessment: Calculate quality metrics for each mesh cell, including orthogonality, smoothness, aspect ratio, and convexity. Orthogonality is defined as the degree to which the interior angles of a cell approach 90°; smoothness is measured by the rate of change of side lengths between adjacent cells; aspect ratio avoids excessive stretching; convexity ensures that the cell is a convex quadrilateral. A weighted sum of these metrics yields a comprehensive quality score, which is then compared to preset thresholds (e.g., orthogonality > 0.8, aspect ratio < 5, etc.), marking cells that fail to meet the standards.

[0143] Optimization and Adjustment: Local adjustments are made to substandard mesh cells. Common operations include: Moving Node Positions: Improving cell shape by adjusting node coordinates through Laplacian smoothing or geometry-based optimization. Local Mesh Refinement: Subdividing poor-quality regions (e.g., dividing a large quadrilateral into four smaller quadrilaterals) to improve local adaptability. Mesh Merging: Merging overly dense or distorted small cells with adjacent cells to improve overall quality.

[0144] Iterative optimization: Repeat the evaluation and adjustment until the quality indicators of all grid cells meet the preset requirements. The number of iterations generally does not exceed 10.

[0145] Mesh Output: Export the optimized structural surface mesh in a standard format (such as CGNS, STL, VTK), and attach metadata such as component semantic tags and boundary condition information for subsequent CFD or structural analysis.

[0146] Example 2;

[0147] Based on Embodiment 1, Embodiment 2 of the present invention provides a detailed description of the intelligent component segmentation in Embodiment 1:

[0148] Embodiment 2 of the present invention provides a method for intelligent component identification and segmentation of aircraft CAD models based on three-modal fusion to achieve intelligent component subdivision. The specific steps include:

[0149] Step S1: Obtain the CAD model of the aircraft to be processed, wherein the CAD model of the aircraft has a boundary representation structure;

[0150] Step S2: Based on the engineering requirements of mesh generation, perform engineering semantic annotation on the faces in the boundary representation structure of the aircraft CAD model to obtain component semantic labels; extract the geometric parameters of each face from the boundary representation structure; generate a multi-view 2D rendering image for each annotated face; associate and store the multi-view 2D rendering image, the geometric parameters, and the component semantic labels to form training samples.

[0151] Step S3: Train a multimodal recognition model using the training samples. The multimodal recognition model is configured to fuse input image features and language semantic information to output semantic labels for components on the surface of the aircraft CAD model.

[0152] Step S4: Input the aircraft CAD model to be identified into the trained multimodal recognition model, identify the surfaces in the aircraft CAD model, and obtain the preliminary surface-part semantic label mapping relationship;

[0153] Step S5: Based on the preliminary face-part semantic label mapping relationship, for faces whose recognition confidence output by the multimodal recognition model is lower than a preset threshold or whose geometric area ratio is less than a set threshold, the geometric rule auxiliary module is activated to perform semantic completion to obtain the final part segmentation result.

[0154] Step S6: Based on the final component segmentation result, extract the key geometric feature lines corresponding to each component from the boundary representation structure of the aircraft CAD model.

[0155] In this embodiment, step S1 limits the processing object to a CAD model with a boundary representation structure, ensuring the feasibility of all subsequent geometric operations (topology analysis, parameter extraction). Steps S2-S3 utilize a multimodal dataset oriented towards mesh engineering semantics and train a specialized model integrating visual, geometric, and linguistic information, enabling the method to understand engineering semantics and replacing manual identification by engineers. Steps S4-S5 employ a collaborative mechanism of AI recognition of large components and geometric rule completion for small features, utilizing AI to handle macroscopic patterns while using geometric rules to ensure detailed integrity, overcoming the limitations of single methods on complex engineering geometries. Steps S6-S7 automatically connect the semantic recognition results to the geometric kernel, extracting key geometric feature lines and converting them into a format directly readable by mesh software, completing the final step from recognition to application and eliminating manual conversion. This solves the end-to-end automation bottleneck in the aircraft CAD mesh generation process, where component recognition relies entirely on manual labor, resulting in extremely low efficiency, inconsistent results, and the inability to directly use the recognition results for subsequent engineering. It achieves a fully automated, high-precision processing flow from the original CAD model to mesh-generated ready engineering data, reducing the time required for manual work, which can take several weeks, to automated processing.

[0156] In step S2, the surfaces in the boundary representation structure of the aircraft CAD model are annotated with engineering semantics. Specifically, in the mesh preprocessing software environment, semantic labels are manually assigned to the surfaces contained in different components based on aerodynamic characteristics or structural functional differences.

[0157] Preferably, the geometric rule auxiliary module determines and corrects the semantic labels of the surface components based on the geometric topological relationships and parametric features of the aircraft CAD model, thereby forming the final component segmentation result.

[0158] Purely data-driven AI models are prone to low confidence or misjudgment when faced with small-sized, weakly textured, feature-fuzzy, or geometrically rare surfaces in the training data (such as slender wing trailing edges, tiny connecting protrusions, and hatches that smoothly transition into the main body). The above approach addresses this by introducing rule modules based on deterministic geometric knowledge to compensate for AI's shortcomings in such cases. General image segmentation models output pixel-level labels, which are difficult to automatically and error-free associate with the precise boundary representation (B-Rep) structure (faces, edges, vertices, and their topology) in CAD models. The above approach directly limits operations to the geometric and topological levels of the original CAD model, ensuring that any judgments and corrections are precisely applied to specific B-Rep surfaces, guaranteeing absolute geometric accuracy. By filling in gaps and correcting errors in the AI ​​recognition results, the final component segmentation results ensure that all geometric details affecting subsequent mesh generation and simulation analysis are covered, reducing the degradation of mesh quality or simulation accuracy caused by missing key small features. This approach eliminates reliance on a single AI model, creating a hybrid intelligence model that combines data-driven (AI) and knowledge-driven (rule-based) approaches. When faced with new aircraft models or those with unique geometries, even if the AI ​​model performs poorly initially, the method can still ensure the correct identification of basic and critical components through geometric rules, thus enhancing the method's generalization ability and practical stability.

[0159] In this embodiment of the invention, the multi-view two-dimensional rendered image is generated as follows:

[0160] For each surface to be processed, perform multi-angle wrapping along its normal vector, render it from each viewpoint, and generate a series of two-dimensional view images.

[0161] Specifically, for each facet, multi-angle rendering is performed along its normal vector. This simulates the process of viewing a geometric facet from various directions, ensuring that the generated image fully reflects the facet's shape characteristics. This solves the problem of generating richly informative 2D image data from a single, planar aircraft CAD model facet, allowing the visual model to learn from it. It provides the visual model with multi-angle, comprehensive observation data, enhancing the model's robustness in recognizing component faces under different viewing angles and lighting conditions.

[0162] In this embodiment of the invention, the training method of the multimodal recognition model is as follows:

[0163] A neural network architecture is constructed, comprising a visual encoder, a language understanding module, and a cross-modal attention fusion module. First, the visual encoding module, based on Vision Transformer (ViT), is responsible for extracting features from the input image and outputting a visual token. Second, the language encoding module, based on a pre-trained large model, is responsible for encoding text prompts into text tokens. Finally, the cross-modal fusion module achieves bidirectional deep interaction between the visual and language tokens through a cross-attention mechanism. After multiple rounds of interaction, the features of the two modalities are aligned and fused in a unified semantic space to form a joint feature representation.

[0164] The neural network architecture is trained under supervision by using the multi-view two-dimensional rendered images and corresponding component semantic label texts in the training samples as input.

[0165] The multimodal recognition model is optimized using a combination loss function, which includes the classification cross-entropy loss function, the segmentation cross-union ratio loss function, and the boundary weighted loss function.

[0166] This paper presents a clear technical framework for fusing multimodal information through an architecture consisting of a visual encoder, a language module, and a cross-modal attention fusion module. Supervised learning is performed using a pre-constructed dataset, enabling the model to learn the mapping from images to semantic labels. A combined loss function is employed to simultaneously optimize classification accuracy, segmentation region consistency, and boundary precision, ensuring the quality of the model's output from multiple dimensions. The paper addresses the core technical path of model implementation by solving the problem of how to specifically construct and train a dedicated recognition model capable of understanding image-semantic associations. An optimized and trained intelligent model capable of accurately associating surface images with component semantic labels has been implemented.

[0167] In this embodiment of the invention, step S4 specifically involves identifying large component surfaces in the aircraft CAD model; step S5 specifically involves semantic completion of local feature surfaces, connecting region surfaces, or low-confidence identification surfaces of the multimodal recognition model in the aircraft CAD model.

[0168] Steps S4 and S5 are defined as a hierarchical collaborative process: the first layer (AI) is responsible for large components (efficiently handling regularities); the second layer (rules) is responsible for local features, connecting surfaces, and other complex situations (accurately handling special characteristics). This design concept is an efficient way to solve engineering complexity problems. It realizes intelligent division of labor and collaboration in the recognition process, significantly improving the completeness of recognizing small and special features while ensuring the efficiency of large component recognition.

[0169] In this embodiment of the invention, the rules for determining the geometric topological relationships and parametric features of the aircraft CAD model include at least one of the following rules:

[0170] Rule 1: If a face is located at the geometric intersection of two identified large component faces, and the area of ​​the face is less than the area of ​​the adjacent large component face, then the component semantic label of the face is determined to be a connecting feature face or an edge face.

[0171] Rule 2: If a surface is directly adjacent to the main wing, and the angle between its normal vector and the normal vector of the main wing's reference plane is greater than or equal to 45°, and this surface is located on the outermost side of the wing span, with an area between 0.5% and 5% of the main wing's area, then it is classified as a winglet or wingtip. The area proportion rule can identify small surfaces at boundaries (such as edges). The direction relationship rule can identify small surfaces that are not parallel to the main airflow direction (such as doors). This provides quantifiable and verifiable rule examples, enabling the system to automatically identify specific types of geometric features such as the wing trailing edge and doors.

[0172] In this embodiment of the invention, the extraction of key geometric feature lines corresponding to each component includes:

[0173] For the components identified by the multimodal recognition model, by analyzing the curvature distribution and boundary topology of its constituent surfaces, a geometric algorithm is used to extract the key geometric feature lines of the components, including its boundary lines and feature contour lines.

[0174] For the feature surfaces identified by the geometric rule assistance module, the corresponding key geometric feature lines are extracted based on their geometric definitions and their relationship with adjacent surfaces.

[0175] For large components identified by AI, geometric algorithms such as curvature and topology analysis are used to automatically extract their boundaries and contours. For small features identified by rules, they are extracted directly based on their geometric definitions (such as the intersection of two faces). Both methods ensure that the output feature lines are accurately associated with the semantic labels of the components. This achieves automated connection from semantic segmentation to geometric feature extraction, outputting geometric data that is directly relied upon for operations such as mesh encryption and block division.

[0176] In this embodiment of the invention, step S5 includes a post-processing step before or after forming the final component segmentation result:

[0177] The semantic labels of components are smoothed based on the adjacency relationship of surfaces to integrate the recognition results of the multimodal recognition model with the supplementary results of the geometric rule auxiliary module, thereby eliminating misjudgments of isolated surfaces; and the final component segmentation results are visualized and exported.

[0178] The label smoothing process is based on the adjacency relationship of faces, performing consistency checks and adjustments on the labels of adjacent faces to eliminate unreasonable abrupt changes or isolated points. The AI ​​results are fused with rule-based results, and visualization and export are supported, forming a complete closed loop from processing to delivery. Post-processing improves the topological consistency and visual smoothness of the segmentation results, and visualization and export functions make the results easy for engineers to verify and use.

[0179] The input to the multimodal recognition model includes natural language instructions; the method allows users to specify the type of part to be recognized or the segmentation requirements by inputting natural language instructions; the visual encoder adopts a Transformer-based architecture, the language understanding module adopts a pre-trained large language model, and the cross-modal attention fusion module adopts a cross-attention mechanism.

[0180] The model's input can include natural language commands, which are then translated into internal tasks that drive the model and geometry engine, achieving intelligent conversion from natural language to concrete operations. This enhances the system's usability and flexibility, allowing users to drive the system to perform specific tasks using natural language (such as segmenting all wing surfaces). The key modules in the architecture are concretized into corresponding selections, ensuring the feasibility of the solution: a visual encoder (ViT or other Transformer architectures), a language understanding module (QwenLM or other large models), and a fusion module (CrossAttn), ensuring the powerful capabilities of each module.

[0181] In this embodiment of the invention, step S7, the structured engineering semantic format includes: a set of component geometric groups for mesh generation software to recognize and call, a set of feature lines describing the key geometric features of the components, and optional mesh generation strategy parameters associated with preset components. By explicitly defining the output format as the structured engineering semantic format required by industrial software, including the component group set, feature line set, and mesh strategy parameters, the output of this method is no longer merely a display result, but rather instructions or data that can be directly read and applied by CAE software, thus achieving a closed loop of engineering value.

[0182] One or more technical solutions provided in Embodiment 2 of the present invention have at least the following technical effects or advantages:

[0183] (1) Full automation of component identification has been achieved, greatly improving the efficiency of mesh preprocessing;

[0184] Because this invention employs a visual-geometric-linguistic three-modal fusion recognition and hierarchical collaborative processing mechanism, it can automatically and accurately complete the identification and semantic segmentation of various aircraft components. The effect is that it shortens the process of identifying medium-sized passenger aircraft components, which originally relied entirely on manual identification and grouping by engineers and took 1-2 weeks, to automated processing, thus solving the primary efficiency bottleneck in the mesh generation process.

[0185] (2) Outputs engineering semantics and geometric features that can directly drive mesh generation;

[0186] Because the output of this invention is not a general image segmentation mask, but rather a face-part label mapping table and key geometric feature lines that are precisely associated with the original B-Rep model, the effect is that the recognition results can be directly converted into components, sets, or layers that mesh generation software can recognize, and key feature lines (such as wing leading / trailing edge lines) used for mesh partitioning and densification control are automatically extracted, achieving recognition-ready functionality and eliminating the large amount of manual conversion and interpretation work required in traditional methods.

[0187] (3) Improved the completeness of identification and engineering practicality of small-sized and weak-featured parts;

[0188] Because this invention employs a layered strategy of AI recognition for large components combined with geometric completion for small features, it intelligently completes small or inconspicuous components such as wing trailing edges and access hatches using geometric rules. The effect is a significant improvement in the overall segmentation integrity, ensuring that subsequent mesh generation covers all necessary geometric details and making the generated mesh more consistent with the geometric integrity requirements of high-fidelity simulation.

[0189] (4) Improve grid quality by supporting smart grid strategy presets through semantic understanding;

[0190] Because the model training of this invention incorporates knowledge of mesh engineering, it possesses engineering semantic understanding capabilities. The effect is that the system can automatically recommend or apply preset mesh generation strategies based on component semantics (such as the leading edge of an airfoil), thereby generating higher-quality meshes more suitable for computation in key aerodynamic regions.

[0191] (5) Possesses strong domain generalization ability and knowledge accumulation value;

[0192] Because the core framework of this invention is not dependent on any specific aircraft model and supports incremental learning, the benefits are significant: only a small amount of data from new aircraft models needs to be labeled and fine-tuned to quickly adapt to new aircraft models, greatly reducing technical maintenance and expansion costs. Simultaneously, the system can distill the mesh generation experience of senior engineers into reusable models and rules, forming core knowledge assets.

[0193] Embodiment 2 of the present invention provides a method for intelligent component recognition and segmentation of aircraft CAD models based on three-modal fusion, the method comprising:

[0194] Step S1 - Model acquisition and format confirmation; Step S2 - Construction of multimodal dataset for engineering semantics; Step S3 - Training of multimodal fusion recognition model; Step S4 - Initial identification of macroscopic components based on AI; Step S5 - Collaborative completion of small features based on geometric rules; Step S6 - Automatic extraction of key geometric feature lines; Step S7 - Structured engineering semantic output.

[0195] This invention provides a fully automated and high-precision component identification and geometric feature extraction capability for mesh generation in computational fluid dynamics and finite element analysis, and is applicable to the digital simulation and manufacturing of complex equipment such as aviation, aerospace, and ships.

[0196] This invention provides a method for intelligent recognition, semantic segmentation, and geometric feature extraction of aircraft components based on a three-modal fusion of vision, geometry, and language. This invention addresses the inefficiencies, inconsistencies, and automation breakpoints caused by reliance on manual component recognition in the aircraft mesh generation process. Specifically, it includes:

[0197] (1) Eliminate the manual dependence in the component identification process and realize the automatic and accurate identification and semantic grouping of each component in the aircraft CAD model; (2) Provide engineering semantic output that can be directly used for mesh generation, including component face sets and key feature lines (such as boundary lines and densification lines); (3) Support the preset of differentiated mesh strategies for different components and realize intelligent mesh planning that is configured upon identification; (4) Improve the automation level and processing efficiency of the entire mesh generation process and provide a reliable pre-processing tool for digital simulation. This invention provides a vision-geometry-language three-modal joint modeling and processing method, the overall architecture of which includes four parts: data construction, model training, inference recognition, geometric processing and mesh strategy generation. The specific steps are as follows:

[0198] (1) Data Construction: A semantic annotation system oriented towards grid requirements:

[0199] 1) B-Rep Surface-Level Semantic Annotation: Using mesh generation software such as NNW-GridStar, and combining experience in aerodynamic and structural mesh generation, surface-level annotations are performed on aircraft CAD models represented in B-Rep format. This includes, but is not limited to: nose, forward fuselage, mid-fuselage, aft fuselage, upper wing surface, lower wing surface, vertical tail, horizontal tail, engine nacelle, etc. After annotation, each component is visualized in a different color for easy verification.

[0200] 2) Multi-view image generation: For each B-Rep face, perform 360-degree multi-view rendering around its normal vector to generate a series of main view images. Scaling, rotation, and lighting adjustments are supported to enhance view coverage and lighting robustness.

[0201] 3) Dataset Augmentation and Construction: Data diversity is enhanced using geometric transformations (rotation, translation, scaling), color perturbation, and noise injection. A structured multimodal dataset is constructed, with each data point including: B-Rep facet ID, multi-view image set, semantic labels, and original geometric parameters.

[0202] (2) Model training:

[0203] 1) Model Architecture Design: ViT (Vision Transformer) is used as the visual encoder to extract image features. CrossAttn (a cross-attention mechanism) is used to align visual features with linguistic features (from QwenLM). QwenLM serves as the language understanding and generation module, responsible for semantic mapping and prompt response.

[0204] 2) Supervised fine-tuning strategy: The input is the main view image of the face, the corresponding semantic label, and the structured prompt words (such as the face belongs to [category]).

[0205] 3) The loss function adopts a ternary combination: semantic cross-entropy loss (optimizes classification accuracy) + IoU loss (improves the overlap of segmented regions) + boundary weighted loss (enhances the accuracy of boundary pixel recognition).

[0206] The loss function is:

[0207] ;

[0208] in, For semantic cross-entropy loss, For IoU loss, For boundary-weighted loss, the weighting coefficients are initially set. , , Adjustments can be made based on subsequent experiments.

[0209] Semantic cross-entropy loss:

[0210] ;

[0211] In the formula, N is the total number of pixels, i.e., the total number of faces in the B-Rep model, C is the number of semantic categories of the component, and y i,c Let p be the true label of the i-th pixel belonging to category c. i,c Let be the probability predicted by the model that the i-th pixel belongs to category c.

[0212] The IoU loss is calculated by averaging the IoU for each part category:

[0213] ;

[0214] Boundary loss ( The initial setting is 0.3 (which can be adjusted based on experiments), where L cls For boundary binary classification loss, L dist Loss for distance transformation:

[0215] ;

[0216] In the formula, This represents the true boundary label of the i-th pixel. This indicates that the pixel is located inside the component. This indicates that the pixel is located on the geometric boundary of two different components; This represents the probability that a pixel predicted by the model is a boundary. Based on the topological relationships of the B-Rep model, when two adjacent faces belong to different components, their shared edge is marked as a boundary.

[0217] Loss for distance transformation:

[0218] ;

[0219] In the formula, The total number of boundary pixels, This represents the distance from pixel i to the nearest real geometric boundary. This represents the distance from pixel i predicted by the model to the boundary. This forces the predicted boundary lines to be closer to the actual geometric boundaries, improving the geometric usability of the segmentation results in CAD models.

[0220] 4) The training employs progressive fine-tuning, first pre-training on large-scale general image-text data, and then fine-tuning on domain image-text pairs labeled with semantic tags for each component of the aircraft. This approach can utilize prior knowledge of general visual language while enabling the model to accurately grasp the fine-grained semantics of the aircraft's unique components, thus improving domain adaptability and recognition accuracy.

[0221] (3) Reasoning and recognition:

[0222] 1) Input: The CAD model of the aircraft to be processed (B-Rep) and natural language prompts (e.g., please identify all fuselage surfaces).

[0223] 2) Layered recognition processing flow:

[0224] The first layer is the overall component recognition: the model extracts visual features face by face, combines them with the semantics of prompt words to make inferences, and prioritizes the recognition of macroscopic large components, such as the entire wing, the entire fuselage, and the tail fin. It outputs the primary semantic label of each B-Rep face to form a face-component mapping table.

[0225] The second layer is small component / feature geometric completion: For small surfaces (such as surfaces belonging to the trailing edge of an wing or local protrusions) whose confidence level is lower than the set threshold or ignored by the model in the previous round of recognition, the geometric rule assistance module is activated instead of relying directly on the AI ​​model for secondary recognition.

[0226] Small parts / features mainly include the following types: Connecting feature surfaces or edge surfaces: Surfaces located at the junction of two large components (such as the upper and lower surfaces of a wing), such as the trailing and leading edges of a wing. Feature surfaces with opening, closing, or maintenance functions: Surfaces surrounded by large components and whose orientation differs significantly from the main body, such as hatches and maintenance access covers. Locally protruding or recessed surfaces: Such as rivet heads, sensor mounts, refueling ports, and other small geometric features. Low-confidence recognition surfaces: Surfaces that the AI ​​model cannot recognize with high confidence due to reasons such as blurred geometric features or insufficient training data coverage.

[0227] 3) Working principle of the geometric rule auxiliary module:

[0228] Input: The set of large components identified by AI (such as the upper surface of the left wing and the lower surface of the left wing) and all the B-Rep surfaces contained therein.

[0229] Processing: Based on the geometric topology and parametric information of the CAD model, preset geometric rules are executed. For example, Rule 1: If a surface belongs to the upper or lower surface of the wing, or is located at the boundary between the two, at the leading edge of the wing, has high curvature, and its spanwise length is close to the entire wing span, it is classified as the leading edge of the wing. Rule 2: If a surface is located at the boundary between the upper and lower surfaces of the wing, and its area is less than 1% of the overall wing area, it is classified as the trailing edge of the wing. Rule 3: If a surface is directly adjacent to the main wing but its normal is significantly deviated from the main wing plane, and it is located at the outermost point of the wing spanwise, with an area between 0.5% and 5% of the main wing area, it is classified as a winglet or wingtip. Rule 4: If a surface is surrounded by a large fuselage component, and its normal is approximately perpendicular to the main fuselage direction, and its area is small, it may be an access hatch or door, and further determination is made based on the geometric characteristics of its adjacent surfaces. Rule 5: If a surface belongs to a large component of the engine nacelle, is located at the foremost end of the nacelle, and has a closed annular boundary and a smooth Gaussian positive curvature transition, then it is classified as an air intake lip.

[0230] Output: Add annotations to the corresponding part categories for facets that conform to the geometric rules, or create new subcategory labels for them.

[0231] 4) Post-processing: Label smoothing is performed based on face adjacency relationships, integrating AI recognition results with geometric rule supplementary results to avoid misjudgment of isolated faces. Component-level result visualization and export are supported.

[0232] (4) Geometric treatment:

[0233] 1) Key Geometric Feature Line Extraction: Based on the complete identification of all components (including large components identified by AI and small features supplemented by geometric rules), boundary lines, curvature extrema lines, and feature contour lines of each component are extracted using B-Rep topology and geometric information. For large components identified by AI, geometric algorithms (such as curvature-based edge detection and topological connectivity analysis) are used for automated extraction. For small features identified by the geometric rule-assisted module, feature lines are extracted directly using their geometric definitions (such as the trailing edge line, i.e., the intersection of the upper and lower wing surfaces), resulting in higher accuracy.

[0234] 2) Output Interface: Feature lines are output in a format supported by NNW-GridStar. Seamless integration with simulation workflows is supported, improving CAE analysis efficiency.

[0235] The visual encoder (ViT) can be replaced by other mainstream networks (such as ConvNeXt, ResNet), and this embodiment of the invention does not impose any restrictions on it. The language model (QwenLM) can be replaced by other large language models (such as GLM, LLaMA), and this embodiment of the invention does not impose any restrictions on it. The specific judgment conditions and parameters of the geometric rules can be adjusted according to the characteristics of different components. The output interface can be adapted to different target mesh generation software (such as ANSA, HyperMesh).

[0236] Example 3;

[0237] Based on Embodiment 1, Embodiment 3 of the present invention provides a detailed description of the polygon subdivision method used in Embodiment 1:

[0238] Among them, the polygon subdivision method can adopt an automatic polygon quadrilateral mesh subdivision method based on deep reinforcement learning, which includes a training phase and an application phase;

[0239] The training phase includes the following steps:

[0240] Step 1: Construct a reinforcement learning environment, initialize the main Q-network and the target Q-network; generate a training dataset containing polygonal geometric sequences with a variable number of vertices, and construct a predefined set of subdivision template actions;

[0241] Step 2: Sample an initial polygon from the training dataset, load the initial polygon into the reinforcement learning environment to initialize its global sub-region set, select a polygon to be processed from the global sub-region set of the reinforcement learning environment according to the preset vertex scheduling strategy, and normalize the variable-length vertex coordinate sequence of the polygon to be processed into a fixed-dimensional state vector through zero-padding and mask marking.

[0242] Step 3: Input the fixed-dimensional state vector into the main Q network, and select the current action to be executed from the predefined set of partitioning template actions based on the ε-greedy strategy;

[0243] Step 4: Feedback the currently executed action to the reinforcement learning environment, which then applies the action to geometrically segment the polygon to be processed, generating several sub-polygons. Quadrilateral sub-regions among the sub-polygons are identified as newly generated quadrilaterals. The sub-polygons are added to the global sub-region set. The polygon to be processed is removed from the global sub-region set to update the global sub-region set. The next polygon to be processed is selected from the updated global sub-region set according to a preset vertex scheduling strategy to form the state vector at the next time step.

[0244] Step 5: Based on the geometric quality, subdivision progress, task completion status, number of execution steps, and legality of the newly generated quadrilateral identified in Step 4, calculate the immediate reward signal;

[0245] Step 6: Combine the current state vector, the current action being executed, the immediate reward signal, the next state vector, and the iteration end label into an experience tuple and store it in the experience replay buffer;

[0246] Step 7: When the batch training conditions are met, randomly sample empirical tuples from the empirical replay buffer;

[0247] Step 8: Calculate the maximum Q-value of the state vector in the next time step in the empirical tuple using the target Q-network, and construct the target Q-value by combining it with the instantaneous reward signal;

[0248] Step 9: Calculate the predicted Q-value of the current state vector in the empirical tuple under the current action using the main Q-network;

[0249] Step 10: Calculate the loss function value between the target Q value and the predicted Q value;

[0250] Step 11: Update the parameters of the main Q-network using the backpropagation algorithm based on the loss function value;

[0251] Step 12: Copy the parameters of the main Q network to the target Q network according to the preset period. Repeat steps 2 to 12 until the training termination condition is met, and save the parameters of the main Q network after training is completed.

[0252] The application phase includes the following steps:

[0253] Step A1: Load the trained main Q-network and construct a reinforcement learning environment for inference;

[0254] Step A2: Obtain the initial polygon to be subdivided, load it into the reinforcement learning environment for inference to initialize the global sub-region set, select the first polygon to be processed according to the preset vertex scheduling strategy, and normalize its variable-length vertex coordinate sequence into a fixed-dimensional initial state vector through zero-padding and mask marking.

[0255] Step A3: Input the initial state vector into the trained main Q network, and select the action with the largest Q value as the current action to be executed;

[0256] Step A4: Perform the current execution action to geometrically segment the polygon to be processed, generate several sub-polygons and update the global sub-region set; select the next polygon to be processed from the updated global sub-region set according to the preset vertex scheduling strategy, and normalize its variable-length vertex coordinate sequence into a fixed-dimensional state vector at the next time step through zero-filling and mask marking;

[0257] Step A5: Repeat steps A3 to A4 until all sub-polygons in the global sub-region set have 4 vertices, and output the final mesh result.

[0258] Existing technologies lack a complete end-to-end automated process for converting arbitrary polygons into full quadrilateral meshes, and the training and application phases are logically disconnected, lacking unified closed-loop control. This method constructs a two-stage architecture including training and application phases. The training phase establishes a reinforcement learning closed loop through steps one through twelve, learning the partitioning strategy; the application phase loads the strategy for inference through steps A1-A5. Specifically, standardized steps are explicitly defined in steps A2 and A4 to ensure that data processing during inference is consistent with that during training. This achieves full automation from data preparation and model training to actual inference. The explicit standardized steps ensure the correct input format of the model in the application phase, avoiding inference failures due to input dimension mismatches and ensuring the system's engineering usability.

[0259] This invention is based on the Deep Q-Network (DQN) framework, modeling the polygon subdivision process as a Markov Decision Process (MDP): State Space: The variable-length vertex coordinate sequence of the polygon to be processed is normalized into a fixed-dimensional tensor (N×3) through zero-padding and masking, serving as the input to the neural network. Action Space: A set of predefined topological segmentation templates for different numbers of vertices (3-6 vertices) (e.g., connecting diagonals, connecting specific vertices) is used, and the agent selects the optimal action from these. Reward Mechanism: A composite reward function is designed. A positive reward is given if an action generates a high-quality quadrilateral; a progress reward is given if the subdivision progress is advanced; a high completion reward is given if the task is completed; and a penalty is given if an invalid action is executed or too many steps are taken. Training and Application: Training Phase: The agent continuously tries and fails in the environment, using experience replay and the target network for stable training, learning the state-action value function (Q-value). Application Phase: The trained model is loaded, and greedy reasoning is performed on new polygons, gradually segmenting them into a full quadrilateral mesh. This paper proposes a reinforcement learning agent to autonomously explore the optimal partitioning path, eliminating reliance on human experience and fixed rules to achieve fully automated partitioning. A state standardization method based on zero-padding and masking mechanisms is proposed, enabling the neural network to uniformly handle polygon inputs with arbitrary numbers of vertices, solving the problem of variable-length inputs. By designing a multi-dimensional composite reward function (quality, progress, completion, and penalty), the agent is guided to generate high-quality quadrilateral meshes and avoid invalid actions, improving mesh quality and success rate. Through end-to-end training, the model can adapt to complex geometries never seen before, achieving generalized partitioning of polygons from simple to complex, thus enhancing generalization ability.

[0260] Preferably, the polygon data in the training dataset is generated in the following way:

[0261] The method employs a random angle-radius method or a convex hull method to generate diverse convex polygon data; or, it acquires the CAD model of the aircraft to be processed and constructs a multimodal feature representation, identifies independent functional components and segmentation boundaries, and outputs structured engineering data containing key geometric feature lines; it parses the key geometric feature lines, reconstructs closed contour curves based on the segmentation boundaries, and discretizes them into vertex sequences to generate polygon data.

[0262] Preferably, the preset vertex scheduling strategy is:

[0263] Traverse the global sub-region set and filter out polygons with more than 4 vertices to form a candidate set;

[0264] If the candidate set is not empty, then the polygons are sorted in descending order of the number of vertices, and the polygon with the most vertices is selected as the next polygon to be processed; if the number of vertices is the same, then the polygon with the lowest geometric quality is selected.

[0265] If the candidate set is empty, the polygon with 4 vertices and the lowest geometric quality is selected as the next polygon to be processed for optimization and adjustment, or the subdivision is determined to be complete.

[0266] In situations where multiple sub-regions exist within a polygon set, randomly selecting the processing order can lead to lengthy meshing paths or even local dead ends (e.g., processing smaller regions first results in larger regions being unsegmentable). This method employs a greedy strategy that prioritizes processing polygons with a high number of vertices (high complexity); if the number of vertices is the same, polygons with lower geometric quality are selected; if no polygon has more than 4 vertices, polygons with 4 vertices of poor quality are optimized or the process is terminated. This approach improves meshing efficiency by prioritizing the most complex regions, avoiding the dilemma of leaving complex regions for last and reducing the total number of meshing steps. It also optimizes global quality by dynamically adjusting the processing order, making the overall mesh generation process more orderly and reducing the global quality degradation caused by suboptimal local selections.

[0267] Preferably, the standardization into a fixed-dimensional state vector is specifically achieved in the following way:

[0268] Set the preset maximum vertex count threshold N max ;

[0269] Get the actual number of vertices n of the polygon currently being processed;

[0270] The construction dimension is (N) max The coordinate matrix of n,2): If n <N max Then, fill the first n rows of the matrix with the two-dimensional coordinates of the n vertices, leaving N vertices. max Subtract n rows and fill with zero vectors; if n≥N max Then calculate the parameters of the intermediate vertices of the polygon to be processed. point, Take any vertex of the polygon to be processed as the starting point, and starting from the starting point, based on the parameters of the intermediate vertices... The middle vertex of the polygon to be processed is determined clockwise. The starting point and the middle vertex are connected to divide the polygon into two sub-polygons. The transformation from mesh to vector is performed based on the sub-polygons.

[0271] The construction dimension is N max Binary mask vector: if n <N max The first n elements of the vector are marked as the first value to represent valid vertices, and the remaining N elements are... max Subtract n elements and label them as the second value to represent the fill data; if n=N max If , then all elements of the vector are marked as the first value.

[0272] In this case, the number of polygon vertices is not fixed (variable-length sequence), while deep neural networks typically require a fixed-dimensional input matrix. Simple truncation will lose geometric information, while simple padding will cause the model to confuse real vertices with the padded data. To address this issue, this invention sets a threshold N. max For less than N max The vertex sequence is zero-padded with a vector, and a binary mask vector is generated simultaneously (1 indicates valid, 0 indicates filled). This achieves a unified input dimension, enabling the network to handle polygons with any number of vertices. Information losslessness and interference resistance: the masking mechanism allows the network to clearly distinguish between real geometric points and filled noise, preventing zero values ​​from being misidentified as the origin, significantly improving the model's accuracy in extracting geometric features.

[0273] Preferably, the execution logic of the ε-greedy strategy is as follows: generate a random number r∈[0,1], if r<ε, then randomly select an action from the subset of legal actions corresponding to the current polygon; if r≥ε, then select the action with the largest Q value output by the main Q network; wherein, the exploration rate ε decreases with the increase of the number of training steps according to a preset decay function.

[0274] In the early and middle stages of deep reinforcement learning training, the main challenge lies in the dilemma between exploration and exploitation: If only exploitation (Greedy) is used, the agent prematurely converges to a local optimum. For example, it might discover that a simple segmentation action (such as always cutting the diagonal) yields a small score, and then perpetually perform this action without trying more complex segmentation strategies that might score higher but perform poorly initially, resulting in mediocre model performance. If only exploration (Random) is used, the agent randomly tries things like a headless fly, leading to extremely low training efficiency, difficulty in convergence, and even an inability to learn basic segmentation logic. The fixed exploration rate problem: If ε remains constant, the agent will still execute random actions with a certain probability in the later stages of training, resulting in policy instability and an inability to output a deterministic optimal solution.

[0275] This method solves the above problems through a dynamic equilibrium mechanism:

[0276] Dual-mode switching: A comparison mechanism between a random number r and a threshold ε is introduced. When r < ε, a legal action is forced to be randomly selected. This ensures that the agent has the opportunity to try actions with temporarily low Q values ​​but unknown potential value (exploring the unknown space). When r ≥ ε, the action with the largest Q value is forced to be selected. This ensures that the agent fully utilizes the knowledge it has learned and executes the optimal strategy (utilizing known experience). Dynamic decay mechanism: ε is defined to decrease monotonically with the number of training steps. In the early stages of training: ε is relatively large (e.g., 0.9), encouraging a large amount of random exploration to quickly cover the action space and collect diverse experience data. In the later stages of training: ε is gradually reduced (e.g., reduced to 0.01), reducing random interference and allowing the agent to focus on fine-tuning and solidifying the optimal strategy, ensuring convergence stability. Legal action constraint: Emphasis is placed on selecting from a subset of legal actions, avoiding geometrically impossible illegal actions during random exploration (e.g., applying hexagonal segmentation rules to triangles), thus improving the effectiveness of exploration. Through sufficient exploration in the early stages, the agent can discover high-value partitioning paths (i.e., action sequences with low short-term rewards but high long-term returns), thereby finding the globally optimal partitioning strategy. As training progresses, the model gradually shifts from a broad-based approach to a more focused one, reducing the interference of invalid random actions on gradient updates and significantly accelerating the convergence speed of the Q-network. The extremely low exploration rate at the end of training ensures the determinism of the model's output, allowing exploration to be directly disabled (ε=0) during the application phase (inference), resulting in stable, reproducible, and high-quality mesh partitioning. Combined with the restriction on the subset of legal actions, this ensures that even data generated by random exploration conforms to geometric topological rules, preventing the experience replay buffer from being contaminated by a large amount of invalid junk data and improving training efficiency.

[0277] Preferred, instant reward signal The calculation method is as follows:

[0278] ;

[0279] in, For quality awards, As a progress reward, To complete the reward, Punishment for inefficiency Error penalties are introduced. A single reward signal can lead to a sparse reward problem, where the agent struggles to receive feedback early in the learning process or focuses solely on completion while neglecting grid quality. This method constructs a composite formula that includes positive rewards for quality, progress, and completion, as well as negative penalties for efficiency and errors. This provides dense and well-guided feedback signals, guiding the agent not only to complete grid partitioning (completion reward), but also to partition well (quality reward), partition quickly (efficiency penalty), and avoid errors (error penalty), thus achieving multi-objective optimization.

[0280] Preferred quality reward The calculation method is as follows:

[0281] The number of newly generated quadrilaterals identified in step 4. ;like Then set ;like ,but ;in, ;in, For orthogonality, For smoothness score, The aspect ratio is used to score. Scoring is based on convexity.

[0282] Preferred, progress reward The calculation method is as follows:

[0283] ;

[0284] in, Increase the reward for the number of quadrilaterals. Reduce rewards to reduce complexity.

[0285] Preferred, completion reward The triggering conditions and calculation methods are as follows:

[0286] Check whether all sub-polygons in the global sub-region set have 4 vertices; if so, then ;in, Rewards for efficiency The sum of quality scores for all quadrilateral meshes; otherwise, set... .

[0287] Preferred, efficiency penalty The calculation method is as follows:

[0288] like ≤10, then ;

[0289] like >10, then ;

[0290] in, Execute the number of steps for the current topology fill action;

[0291] Error Penalty The calculation method is as follows: ;in, Penalty for abnormal number of sub-regions Penalty for invalid actions.

[0292] One or more technical solutions provided in Embodiment 3 of the present invention have at least the following technical effects or advantages:

[0293] High automation and robustness: It can process polygons with any number of vertices without human intervention, significantly reducing the labor cost of preprocessing.

[0294] Excellent mesh quality: The generated quadrilaterals perform well in terms of orthogonality, smoothness, aspect ratio, etc., reducing the error of subsequent simulation calculations.

[0295] High convergence speed: Through a reasonable scheduling strategy (prioritizing polygons with a large number of vertices) and reward guidance, the number of steps required for subdivision is significantly reduced.

[0296] Flexible data compatibility: The proposed standardization method perfectly resolves the contradiction between variable-length geometric data and fixed neural network input, ensuring lossless information transmission.

[0297] Example 3 provides an automatic polygon quadrilateral mesh generation method based on deep reinforcement learning. The complete process of automatic polygon quadrilateral mesh generation by this method includes the following steps:

[0298] 1) Environment initialization: Generate or input the polygon dataset to be subdivided, and initialize the reinforcement learning environment;

[0299] 2) State observation construction: For the polygon to be processed, construct state observations containing geometric information and effective masks;

[0300] 3) Agent decision-making: The DQN agent selects the optimal partitioning template action based on the current state;

[0301] 4) Action execution and state transition: Apply the selected topology fill template to subdivide the polygon and update the sub-region set;

[0302] 5) Reward Calculation: Calculate immediate rewards based on the mesh quality after topology filling;

[0303] 6) Experience storage: Store state transition experience to optimize the agent's policy network;

[0304] 7) Batch training: The DQN learning algorithm is used to train the main Q-network;

[0305] 8) Target Q-network update: Periodically update the target Q-network and copy the parameters of the master Q-network to the target Q-network;

[0306] 9) Model Deployment: Apply the trained optimal strategy to the new polygon subdivision task.

[0307] The core principle of Example 3 is to model the polygon mesh subdivision process as a Markov Decision Process (MDP): 1) State space: geometric feature representation of the polygon (vertex coordinates + effective mask); 2) Action space: a predefined set of subdivision templates (9 subdivision operations); 3) Reward function: a reward signal based on mesh quality to guide the agent to learn a high-quality subdivision strategy; 4) State transition: geometric shape changes caused by template application; through deep Q-learning, the agent learns the mapping function from polygon state to subdivision action to maximize long-term cumulative reward.

[0308] Detailed explanation of each step:

[0309] Step 1: Environment Initialization: 1) Execution Entity: Training System; 2) Execution Content: Generate diverse convex polygon training datasets; Define Q-network; Define template actions and initialize core components; 3) Specific Implementation:

[0310] Polygon generation methods include: random generation methods such as the angle-radius method (simple and efficient, ensuring convexity, but with relatively regular shapes) and the convex hull method (ensuring strict convexity, resulting in more natural and diverse shapes); and geometric constraint-based methods such as regular polygon generation and star-shaped polygon generation (there are many automatic polygon generation algorithms available, and the choice can be made based on the specific application). In addition, initial polygons can be generated by recognizing geometric components from a digital model using methods provided by the intelligent component recognition and segmentation method for aircraft CAD models based on three-modal fusion.

[0311] Q-network definition (the main / target network structures are exactly the same): Input: A straight-edge polygon represented by an ordered sequence of vertices; Output: The execution probability of all topology-filling template actions; Model structure: The model is a classification network (only a deep learning neural network reference template is provided here, which can be modified according to the actual situation), as follows. Figure 7 As shown, Figure 7 This is a schematic diagram of the Q-network model structure.

[0312] The model consists of two 1D convolutional layers (the neural network processes the sequence of polygon vertices (extracting meaningful local geometric features from the original vertex coordinates to provide high-quality input for subsequent LSTM and attention mechanisms)) and two normalization layers; one LSTM layer (enabling the neural network to understand the sequential relationships between vertices, a key capability for processing sequential geometric data like polygons); one multi-head attention layer (multi-head attention allows the model to adaptively focus on the most important parts of the polygon); three fully connected layers (the three fully connected network converts the rich features extracted by the previous layers into the final classification decision); and two Dropout layers to prevent overfitting.

[0313] Template action definition:

[0314] as follows Figures 3 to 6 As shown, nine template actions are defined to cover the topology filling method from triangle to hexagon.

[0315] like Figures 3 to 6 As shown, this invention predefines nine basic topology filling templates. These templates are categorized based on the number of polygon vertices N, and are designed to transform the current polygon into a combination containing at least one quadrilateral subregion, or to complete the final subdivision, through a single geometric segmentation operation. Category 1: Processing template for triangles (N=3); Category 2: Processing template for quadrilaterals (N=4); Category 3: Processing template for pentagons (N=5); Category 4: Processing template for hexagons (N=6). These nine templates represent the minimum complete set selected through geometric topology analysis.

[0316] These templates compress an infinite number of geometric cutting possibilities into nine discrete options, transforming the action space of reinforcement learning from continuous high-dimensional to low-dimensional discrete, improving training efficiency by several orders of magnitude. This allows the model to learn complex partitioning strategies within a finite number of steps, accelerating convergence. Predefined templates physically prevent the generation of illegal meshes (such as self-intersecting or non-manifold edges). The agent can only choose from legal options, eliminating the need for complex penalty terms in the reward function to correct geometric errors and ensuring topological legality. These nine templates encapsulate the geometric heuristics from traditional mesh partitioning algorithms (such as paving and mapping methods). They incorporate expert knowledge, achieving a combination of data-driven and rule-driven approaches. While the templates are fixed, the template selection strategy (Q-network) is adaptive. Faced with different polygon shapes (flat, elongated, concave), the network automatically selects the most suitable template from these nine options, achieving robust partitioning of arbitrarily complex geometries with high adaptability.

[0317] Initialize core components:

[0318] Initialize the main Q-network (keeping parameters continuously updated) and the target Q-network (with fixed parameters, periodically synchronizing with the main Q-network), keeping the main Q-network and the target Q-network consistent during initialization; initialize the optimizer; initialize the experience replay pool; define hyperparameters (discount factor, ε-greedy, batch training size, number of update steps for the target Q-network, total number of training steps, etc.).

[0319] Step 2: State observation construction:

[0320] 1) Execution subject: State extraction module; 2) Execution content: Standardize variable-length polygon data into a fixed-dimensional state representation; 3) Specific implementation:

[0321] Polygon coordinates fill: fills the vertex coordinates to a fixed length (6 vertices); valid mask creation: identifies the actual vertex positions and distinguishes the fill data; status format: {"polygon":(6,2) coordinate array, "mask":(6) mask array}.

[0322] The core objective of filling is to standardize the coordinates of a polygon with any number of vertices (≤6) into an array of 6 fixed vertices, each containing a two-dimensional x / y coordinate, and to distinguish between real vertices and filled vertices through a mask, ultimately outputting a state representation with a unified dimension.

[0323] The specific implementation method is as follows:

[0324] Polygon coordinates fill (polygon array: shape=(6,2));

[0325] Preprocessing:

[0326] Coordinate normalization:

[0327] To eliminate the influence of absolute coordinate scale (such as differences in coordinate range between different polygons), the vertex coordinates of the original polygons are first normalized using the industry-standard Min-Max Scaling method, as shown in the following formula:

[0328] ;

[0329] in, and These are the vertices of the current polygon before and after normalization, respectively. value, and These are the vertices of the current polygon to be normalized. and Axis coordinates and Each of the vertices of the current polygon The maximum and minimum values ​​of the axis coordinates. and Each of the vertices of the current polygon The maximum and minimum values ​​of the axis coordinates;

[0330] Calculation logic:

[0331] (1) Traverse all vertices of the current polygon and extract all vertices. Minimum value of coordinates Maximum value Similarly, extract coordinates , ;

[0332] (2) If (e.g., a vertical line segment), then (To avoid a denominator of 0); the same applies to the y-axis;

[0333] (3) After normalization, the coordinate range is fixed at [0,1] to ensure the consistency of scale of different polygons.

[0334] Vertex filling and arrangement:

[0335] Rules: Preserve the original vertex order (clockwise). If there are fewer than 6 vertices, fill the end of the original vertex list with invalid vertices. If the number of vertices is greater than 6, segment the polygon before filling and arranging the vertices (calculate the parameters of the intermediate vertices of the polygon to be processed). , Take any vertex of the polygon to be processed as the starting point, and starting from the starting point, based on the parameters of the intermediate vertices... Determine the middle vertex of the polygon to be processed clockwise, and connect the starting point and the middle vertex to divide the polygon into two sub-polygons.

[0336] Invalid vertex values: uniformly fill (0.0, 0.0).

[0337] Example (with) Figure 8 (Taking polygon 1, which is the result of dividing a heptagon, as an example)

[0338] Original vertices (normalized): [(0.1,0.2),(0.3,0.4),(0.5,0.6),(0.7,0.8),(0.9,0.1)] (5 vertices) Polygon array after filling: [

[0340] [0.1,0.2],#True vertex 1

[0341] [0.3,0.4],#True Vertex 2

[0342] [0.5, 0.6], #True Vertex 3

[0343] [0.7,0.8],#True Vertex 4

[0344] [0.9, 0.1], #True Vertex 5

[0345] [0.0,0.0], # Fill vertex 1 ];

[0347] Dimension Explanation: 6 represents a fixed number of vertices, 2 represents the x / y coordinates of each vertex, and the final polygon is a 6-row, 2-column two-dimensional array (shape=(6,2)).

[0348] Valid mask creation (mask array: shape=(6,)):

[0349] The mask array has the same length as the number of vertices. 1 indicates a real vertex and 0 indicates a filled vertex. Only the validity needs to be distinguished; normalization is not required.

[0350] Example (corresponding) Figure 8 Polygon 1 in the middle):

[0351] mask array: [1,1,1,1,1,0]

[0352] Dimension Explanation: A 6-dimensional one-dimensional array, where each element is an integer of type 0 or 1.

[0353] Final state format:

[0354] {

[0355] "polygon":np.array([[0.1,0.2],[0.3,0.4],[0.5,0.6],[0.7,0.8],[0.9,0.1],[0.0,0.0]]),

[0356] "mask":np.array([1,1,1,1,1,0])

[0357] };

[0358] For specific implementation methods, please refer to the literature (min-max normalization): Data Mining: Concepts and Techniques (Third Edition) Pages 83-124. The embodiments of the present invention will not be elaborated or limited accordingly.

[0359] Step 3: Agent Decision Making

[0360] Execution subject: DQN agent; Execution content: Selecting partitioning actions based on the ε-greedy policy; Decision mechanism: Exploration phase (when the random probability is less than ε-greedy): Randomly select the template action for the current point to expand the search space; Utilization phase (when the random probability is greater than ε-greedy): Select the optimal action with the largest Q value (the template action with the highest score); ε decay: Gradually reduce the exploration probability as training progresses.

[0361] Step 4: Action Execution and State Transition

[0362] Execution subject: Template action execution module; Execution content: Apply the selected template to subdivide the current polygon; Key technology: The template library contains 9 subdivision operations for different numbers of vertices to ensure geometric validity.

[0363] For details on the nine templates in the template library, please refer to Table 1, which is a template information table.

[0364] Table 1

[0365] template Target graphics Connection method (geometric definition) Generating sub-regions Technical effects / applications Template 1 (e.g.) Figure 2 Triangle template triangle 1. Connect the midpoints of the two legs of the triangle to obtain a line segment parallel to the base; 2. Draw a vertical line segment from the midpoint of this line segment to the midpoint of the base. One small triangular region + two congruent trapezoidal regions, for a total of three sub-regions. Achieving triangular symmetric subdivision facilitates mesh generation, uniform region partitioning, and maintains geometric topological consistency. Template 2 (e.g.) Figure 3 Left 1) Quadrilateral non-uniform template quadrilateral 1. Draw line segments from the left and right vertices of the upper base of the quadrilateral to the left and right 1 / 3 points of the lower base, respectively; 2. Draw vertical line segments upward from the left and right 1 / 3 points of the lower base, intersecting the left and right line segments from step 1 at a single point inside; 3. Connect the left and right internal points from step 2. The four sub-regions are: a small rectangle on the left, a trapezoid in the upper middle, a square in the lower middle, and a small rectangle on the right. Divide the quadrilateral into equal parts along the horizontal direction to adapt to non-uniform layout and modular area allocation scenarios. Template 3 (e.g.) Figure 3 Left 2) Quadrilateral center nested template quadrilateral 1. Draw line segments from the left and right vertices of the upper base of the quadrilateral to the left and right 1 / 3 points of the lower base, respectively; 2. Draw line segments from the left and right vertices of the lower base of the quadrilateral to the left and right 1 / 3 points of the upper base, respectively; 3. Connect the four interior points from steps 1 and 2 in sequence. A small central square plus four surrounding trapezoids, totaling five sub-regions. Achieve centrally symmetric subdivision of quadrilaterals, suitable for uniform mesh generation. Template 4 (e.g.) Figure 3 Right 1) Quadrilateral asymmetric template quadrilateral 1. Draw a line segment from the left vertex of the upper base of the quadrilateral to the right half-point of the lower base; 2. Draw a vertical line segment upward from the midpoint of the lower base to the point inside the quadrilateral in step 1; 3. Draw a horizontal line segment to the left from the midpoint of the right base to the point inside the quadrilateral in step 1. There are three sub-regions: a trapezoid at the bottom left, a trapezoid at the top right, and a square at the bottom right. It provides asymmetrical yet regular quadrilateral partitioning to meet the functional needs of differentiated areas. Template 5 (e.g.) Figure 3 Right 2) Quadrilateral division template quadrilateral Connect the horizontal and vertical midlines of the square. Four congruent small quadrilaterals, forming four sub-regions. The basic quadrilateral template is suitable for general scenarios such as coordinate partitioning, uniform meshing, and pixelated region splitting. Template 6 (e.g.) Figure 4 Left 1) Pentagonal diagonal template pentagon Connect any vertex to a non-adjacent vertex (diagonal). One triangular region + one quadrilateral region, for a total of two sub-regions. Decomposing the pentagon into its basic geometric shapes facilitates geometric subdivision and simplification. Template 7 (e.g.) Figure 4 Right 1) Pentagonal symmetrical template pentagon The vertical line segment connecting the upper vertex and the midpoint of the lower base. Two symmetrical quadrilateral regions, comprising two sub-regions. Maintaining the lateral symmetry of the pentagon is suitable for geometric simplification and symmetry analysis. Template 8 (e.g.) Figure 5 Left 1) Hexagonal horizontal template hexagon Connect the midpoints of opposite sides in the horizontal direction (horizontal median) There are two isosceles trapezoidal regions, one above the other, making a total of two sub-regions. The hexagon is symmetrically divided horizontally, maintaining the same shape and area between the upper and lower regions. Template 9 (e.g.) Figure 5 Right 1) Hexagonal vertical template hexagon Connect the midpoints of opposite sides in the perpendicular direction (perpendicular median) There are two pentagonal regions on the left and right, making a total of two sub-regions. The hexagon is symmetrically divided along the vertical direction, keeping the shape and area of ​​the left and right regions consistent.

[0366] Step 5: Reward Calculation

[0367] Execution Entity: Reward Calculation Module; Execution Content: Comprehensive evaluation of segmentation quality; Reward Design:

[0368] The reward design formula is shown below. For immediate reward signals, For quality awards, As a progress reward, To complete the reward, Punishment for inefficiency Punishment for mistakes.

[0369] ;

[0370] Basic Quality Bonus: The basic quality bonus is only applied to quadrilateral sub-regions; no quality bonus is awarded if no quadrilaterals are generated. The formula for the basic quality bonus is as follows: The sum of the quality scores for all quadrilateral grids, the number of newly generated quadrilaterals identified in step 4, i.e. the number of valid sub-regions (quadrilateral sub-regions), is multiplied by 10 to amplify the reward magnitude.

[0371] ;

[0372] in The calculation formula is as follows. Where... Orthogonality (angles close to 90 degrees score higher) The smoothness score is awarded based on the smoothness of the edges (those with a gentle change in the length of adjacent edges receive a higher score). Score the aspect ratio (avoid overstretching of quadrilateral sub-regions). Scoring is based on convexity (ensuring the quadrilateral is a convex polygon). 0.4, 0.3, 0.2, and 0.1 are respectively... The weighting percentage.

[0373] ;

[0374] Progress reward: Quantitatively evaluates the progress made by a single partitioning action, providing intermediate guidance signals for the reinforcement learning agent and avoiding rewarding only upon final success (the sparse reward problem). The progress reward formula is as follows, where... Increase the reward for the number of quadrilaterals. To reduce complexity, the reward is reduced (the average number of vertices reduces the reward). Versions 2.0 and 1.5 are respectively... The weighting percentage.

[0375] ;

[0376] in The calculation formula is as follows. This represents the current number of quadrilateral subregions. The number of quadrilateral subregions before topological filling.

[0377] ;

[0378] and The calculation formula is as follows. This represents the average number of vertices in all sub-regions before topology filling. This represents the average number of vertices in all sub-regions after topological filling.

[0379] ;

[0380] Completion Reward: A completion reward is awarded when all sub-regions are filled with quadrilaterals. The completion reward formula is as follows, where... There are efficiency rewards (the fewer steps, the higher the reward). The base score is 20.0, calculated as the sum of the quality scores for all quadrilateral grids. 2.0 and 5.0 are respectively... The weighting percentage.

[0381] ;

[0382] in The calculation formula is as follows. The total number of steps performed for the topology filling action.

[0383] E bon =max(10-Step t ,0);

[0384] Efficiency penalty: An efficiency penalty is applied based on the total number of steps taken in the topology-filling action. The efficiency penalty formula is as follows, where... The number of steps required to perform the current topology filling action, with 0.1 representing the step penalty weighting. The penalty is divided into a linearly increasing penalty and an overstep penalty. When the maximum number of steps (10) is exceeded, the efficiency penalty increases by 5.0.

[0385] ;

[0386] ;

[0387] Error penalty: A penalty for invalid or erroneous operations. The formula for error operation penalty is as follows, where... Penalize invalid actions (such as applying a quadrilateral template to a triangle). Penalize for abnormal number of sub-regions.

[0388] ;

[0389] ;

[0390] ;

[0391] Step Six: Experience Storage;

[0392] Execution subject: DQN learning algorithm; Execution content: Storing experience into the replay pool; Stored content:

[0393] experience tuples Store in the return tank. This represents the current state (the coordinates of all vertices of the sub-regions after the current polygon topology is filled). The topology fill template action selected for the current step. The reward value obtained after performing topology filling on the current state. This refers to the next stage state after performing the topology fill action (the coordinates of all sub-region vertices after performing polygon topology fill). This is the end label for iterative execution.

[0394] Step 7: Batch Training

[0395] Execution subject: DQN learning algorithm; Execution content: Batch training optimization through random sampling of the experience replay pool and deep Q-learning of the target network mechanism; Specific implementation:

[0396] Random sampling of the experience replay buffer: The experience replay buffer serves as the agent's memory, storing a large amount of state transition experience. During each training session, a small batch of experience data is randomly sampled from this buffer for learning.

[0397] Optimal Future Reward Estimation of the Target Q-Network: The target Q-network is a lagged replica of the main Q-network, used to calculate a stable estimate of the future reward. First, the next state is... The input is fed into the target Q-network to obtain the Q-value predictions for all possible topology-filling actions. Then, for each next state, the action with the largest Q-value is selected, representing the optimal future reward for that state. Finally, the largest Q-value is multiplied by a discount factor. , represents the present value of future rewards. For states that cause the task to terminate, the future rewards are zero, and only the immediate rewards are calculated.

[0398] Target Q value calculation:

[0399] The calculation formula is as follows, where For the target Q value, Stored in the experience pool of the main Q network Reward value in the state, As a discount factor, For the target Q network in The probability of executing the optimal topology filling template action under the given state.

[0400] ;

[0401] Calculation of the current Q-value of the main Q-network:

[0402] First, set the current Input the main Q-network, obtain Q-value estimates for all actions through the network's forward propagation, and then use the gather operation to select the Q-value corresponding to the actual action to be performed from all the Q-values. .

[0403] Calculation of time-series differential loss:

[0404] Huber loss function Using Smooth L1 Loss provides better robustness to outliers and is more stable for training compared to mean squared error; temporal difference error. : Calculates the difference between the current Q-value and the target Q-value, reflecting the accuracy of the agent's prediction; regularization term :Add to Regularization prevents overfitting and penalizes excessively large network weights. The specific formula is as follows.

[0405] ;

[0406] Backpropagation and parameter optimization:

[0407] The gradient of the network parameters with respect to the loss is calculated through backpropagation of the loss function.

[0408] Step 8: Target Q network update;

[0409] 1) Execution subject: main Q network, target Q network; 2) Execution content: when the preset target Q network update step number is reached, synchronize the main Q parameters to the target Q network.

[0410] Step 9: Target Q network update;

[0411] 1) Execution subject: The optimal policy network after training;

[0412] 2) Execution Content and Target: Fix the parameters of the optimal policy network after training convergence. For a new geometric model, load this network and execute steps two through six for it. The agent can then quickly output an optimized action sequence based on the learned policy.

[0413] Logical Relationship: This step demonstrates the ultimate value and generalization ability of the method, enabling automated optimization that is ready to be trained and used immediately.

[0414] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0415] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An automated method for generating structural surface meshes based on topology filling, characterized in that, The method includes: Step 1: Obtain the 3D CAD model to be processed, perform component identification and subdivision on the 3D CAD model, and obtain multiple independent components and their geometric contour data; Step 2: According to the preset component adjacency order, perform ordered topology filling on each component sequentially, including: based on geometric contour data, abstract the geometric contour of the current component to be processed into a polygon, and obtain the number of first vertices of the filled components adjacent to the current component on the intersecting edge; adjust the number of second vertices of the current component on the intersecting edge based on the number of first vertices, so that the number of second vertices is the same as the number of first vertices; use a predefined topology filling template to recursively subdivide the adjusted polygon until all sub-regions of the current component are quadrilaterals, and generate the topology line of the current component; Step 3: Project the generated topology lines onto the surface of the 3D CAD model, perform geometric fit calibration, and obtain the projected topology lines; Step 4: Perform adaptive interpolation processing on the projected topology lines. Based on the surface curvature of the 3D CAD model and the preset simulation calculation conditions, dynamically interpolate to generate point cloud data on the projected topology lines. Step 5: Generate an initial structural surface mesh based on the point cloud data corresponding to each quadrilateral sub-region; Step 6: Project the nodes of the initial structural surface mesh onto the surface of the 3D CAD model, optimize the quality of the projected mesh, and generate the final structural surface mesh.

2. The method for automatically generating structural surface meshes based on topology filling according to claim 1, characterized in that, Step 1 specifically includes: Step 1.1: Obtain the 3D CAD model to be processed, wherein the 3D CAD model has a boundary representation structure; Step 1.2: Based on the engineering requirements of mesh generation, perform engineering semantic annotation on the faces in the boundary representation structure of the 3D CAD model to obtain component semantic labels; extract the geometric parameters of each face from the boundary representation structure; generate a multi-view 2D rendering image for each annotated face; associate and store the multi-view 2D rendering image, the geometric parameters, and the component semantic labels to form training samples; Step 1.3: Train a multimodal recognition model using the training samples. The multimodal recognition model is configured to fuse input image features and language semantic information to output semantic labels for components on the surfaces of a 3D CAD model. Step 1.4: Input the 3D CAD model to be identified into the trained multimodal recognition model, identify the faces in the 3D CAD model, and obtain the preliminary face-part semantic label mapping relationship; Step 1.5: Based on the preliminary face-component semantic label mapping relationship, for faces whose recognition confidence output by the multimodal recognition model is lower than a preset threshold or whose geometric area ratio is less than a set threshold, the geometric rule auxiliary module is activated to perform semantic completion to obtain the final component segmentation result. Step 1.6: Based on the final component segmentation result, extract the key geometric feature lines corresponding to each component from the boundary representation structure of the 3D CAD model. The component geometric contour data includes the final component segmentation result and the key geometric feature lines.

3. The automated method for generating structural surface meshes based on topology filling according to claim 2, characterized in that, After step 1 and before step 2, the following is also included: Based on the semantic tags of the components and combined with a preset rule base, the component generation order of the whole machine surface mesh is determined, resulting in one or more component generation sequences.

4. The automated generation method for structural surface mesh based on topology filling according to claim 3, characterized in that, Step 4 includes: The component to be generated is determined according to the component generation sequence. The topological structure block, semantic label and working condition parameters of the component to be generated are input into the trained graph neural network (GNN) model. The GNN model collaboratively predicts the size distribution function of all topological edges and outputs the initial node distribution of each topological edge. The topological structure block is composed of the topological line generated in step 2 and the key geometric feature line obtained in step 1. For subsequent components in the component generation sequence, the node coordinate sequence on the shared topological edge is inherited from the generated adjacent component mesh as a rigid constraint. Based on the rigid constraint and the size distribution function predicted by the graph neural network (GNN) model, the node distribution of the unconstrained edge is determined. The interpolation point density is adjusted to generate point cloud data in combination with the surface curvature and preset simulation calculation conditions.

5. The method for automatically generating structural surface meshes based on topology filling according to claim 1, characterized in that, Step 2 uses a predefined topology filling template to recursively subdivide the adjusted polygon, including: The variable-length vertex coordinate sequence of the current polygon to be processed is normalized into a fixed-dimensional state vector through zero-padding and mask marking; The fixed-dimensional state vector is input into the trained deep reinforcement learning model. The deep reinforcement learning model selects the current action to be executed from a predefined set of topological filling template actions that cover triangles to hexagons based on a preset vertex scheduling strategy. The current action is then applied to geometrically segment the polygon to be processed, generating several sub-polygons and identifying quadrilateral sub-regions within them.

6. The method for automatically generating structural surface meshes based on topology filling according to claim 1, characterized in that, In step 2, the number of first vertices of the filled components adjacent to the current component on the intersecting edge is obtained; the number of second vertices of the current component on the intersecting edge is adjusted based on the number of first vertices, specifically as follows: Read the vertex coordinate sequence of the intersection edge between the filled component and the current component, and count the number of vertices as the first vertex count; expand or reduce the vertex count of the intersection edge corresponding to the current component from the initial value to be equal to the first vertex count, and recalculate the adjusted coordinate position of each vertex based on linear interpolation or geometric feature resampling algorithm to realize the one-to-one correspondence between points on the intersection edge.

7. The method for automatically generating structural surface meshes based on topology filling according to claim 1, characterized in that, Step 4 involves adaptive interpolation processing of the projected topology lines, specifically including: The curvature distribution of the surface of the 3D CAD model corresponding to the projected topology line is calculated. The interpolation point density is increased in the region where the curvature is greater than the first preset threshold, and the interpolation point density is reduced in the region where the curvature is less than the second preset threshold. Based on the working parameters of the subsequent simulation calculation, the minimum size threshold and the maximum size threshold of the interpolation points are adjusted so that the generated mesh size meets the calculation accuracy requirements.

8. The method for automatically generating structural surface meshes based on topology filling according to claim 1, characterized in that, Step 6 involves optimizing the quality of the projected mesh, specifically including: Calculate the orthogonality, smoothness, aspect ratio, and convexity quality indices of the grid cells, obtain the index calculation results, and evaluate whether each grid cell meets the preset quality threshold based on the index calculation results; For mesh cells that do not meet the preset quality threshold, adjustments are made by moving node positions, refining local meshes, or merging them. Repeat the evaluation and adjustment process until the quality indicators of all grid cells meet the preset quality threshold.

9. The method for automatically generating structural surface meshes based on topology filling according to claim 2, characterized in that, The geometric rule auxiliary module determines and corrects the semantic labels of the face components based on the geometric topological relationships and parametric features of the 3D CAD model, thus forming the final component segmentation result.

10. The method for automatically generating structural surface meshes based on topology filling according to claim 1, characterized in that, Step 5, which generates the initial structural surface mesh, specifically includes: For each quadrilateral sub-region, based on the point cloud data on its boundary, uniformly distributed grid nodes are generated inside the sub-region using interpolation; the grid nodes are connected to form quadrilateral grid cells, and the vertex index and coordinate information of each grid cell are recorded.

Images