A method for automatically generating RTM grid based on implicit distance mapping and normal growth

Abstract

This application provides an automatic RTM mesh generation method based on implicit distance mapping and normal growth, comprising: acquiring a triangular mesh on the surface of a composite material part; obtaining a local normal field on the part surface by calculating the unit normal vector of each triangular facet; using the implicit distance field as the geometric representation basis, converting the layup thickness into a distance offset value according to the thickness mapping relationship to generate an enhanced initial reference surface; projecting points on the part surface onto the implicit distance field through the thickness mapping relationship, calculating the signed distance value of each point, determining the nested region affiliation, and obtaining supplementary nested space affiliation labels; using the optimized material region as boundary constraint conditions, performing voxel growth operations layer by layer along the local normal field of each layer to generate an initial volume mesh corresponding to each layup; and outputting the final solid mesh for simulation analysis of resin transfer molding process flow and curing.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an automatic generation method for RTM meshes based on implicit distance mapping and normal growth. Background Technology

[0002] In the process of generating solid meshes for composite variable-angle plywood components, the core technical challenge lies in how to effectively integrate the discrete normal field of the triangular mesh on the outer surface of the part with the implicit distance field generated by the layer-by-layer offset of each ply, so as to accurately realize the nested division of material regions and the construction of continuous meshes.

[0003] The specific problem is that due to the drastic change in curvature and the discretization error of the triangular mesh, the local normal field on the surface of the part is easily offset along the normal direction after the reference surface of the adjacent ply is shifted. This can easily lead to self-intersection, wrinkles, or deviation from the true geometry in the concave and convex areas and where the thickness accumulates. Meanwhile, although the implicit distance field can provide a continuous and smooth offset expression, when mapping discrete part surface points to the implicit field to determine nesting, due to the ambiguity of the mapping method, sampling accuracy and symbol distance, the cell center point is often incorrectly assigned to the unexpected ply area, causing local confusion of nesting labels. Furthermore, these erroneous initial attribution labels will spread to the surrounding areas through the region growth correction process, leading to increased contradictions in the labels of adjacent units, distortion of boundary regions, and ultimately causing severe overlap, gaps, or topological breaks in the interlayer of subsequent voxel solid units grown along the normal direction, which cannot meet the stringent requirements of geometric closure and material continuity for RTM process simulation.

[0004] The above problems are closely coupled and all stem from the logical inconsistencies and accuracy losses between the discrete normal field and the continuous implicit distance field in the geometric expression, spatial mapping, and region attribution determination stages.

[0005] In the prior art, CN114548527A discloses a method for generating composite material layup meshes, which only uses normal offset to generate layup meshes without incorporating implicit distance fields, easily leading to mesh self-intersections; CN115661558A discloses an RTM process mesh generation method, which uses voxelization but does not optimize the nested space attribution labels, easily resulting in interlayer overlap and gaps; the journal article "Research on RTM Process Mesh Generation Technology" published in the 2023 issue of the Journal of Composite Materials (Vol. 40, No. 5) only targets regular curved surface parts and cannot adapt to the variable angle layup design of complex curved surfaces. This application addresses the shortcomings of the above-mentioned prior art by integrating implicit distance mapping and normal growth, determining the attribution labels of double-layer nested spaces, and implementing a three-level mesh repair process, thereby achieving high-quality automatic mesh generation for complex curved surface variable angle layup parts. Summary of the Invention

[0006] This invention provides an automatic RTM mesh generation method based on implicit distance mapping and normal growth, mainly including: Obtain the triangular mesh on the surface of the composite material part, and calculate the unit normal vector of each triangular facet to obtain the local normal field of the part surface; For a variable angle ply sequence, starting from the bottommost ply, the initial reference surface of the ply is obtained by offsetting along the normal direction based on the local normal field according to the ply angle and ply thickness. Using an implicit distance field as the basis for geometric representation, the ply thickness is converted into a distance offset value according to the thickness mapping relationship to generate an enhanced initial reference surface; Perform a spatial point inclusion test on the initial reference surface and the triangular mesh on the part surface to determine the ply area to which the center point of each triangular facet in the triangular mesh on the part surface belongs, and obtain a preliminary nested spatial attribution label. By projecting the part surface points into the implicit distance field through the thickness mapping relationship, calculating the symbolic distance value of each point, determining the nested region affiliation, and obtaining the supplementary nested space affiliation label; Based on the initial nested space attribution labels, the region growing algorithm is used to correct the incorrectly assigned regions, and the nested space is re-divided through the consistency constraint of adjacent unit labels to obtain the optimized material region division result; Using the optimized material region as the boundary constraint, voxel growth operations are performed layer by layer along the local normal field of each layer to generate the initial volume mesh corresponding to each layup. Interface overlap and gap detection is performed on the initial volume mesh between adjacent layers to determine whether there are cell overlaps or gaps. The cells at the interface are made to fit together by adjusting the local mesh point displacement to obtain an intermediate volume mesh without overlap or gaps. For the topologically inconsistent regions that still exist in the intermediate mesh, a voxel repair algorithm is used to obtain a geometrically closed and materially continuous final solid mesh by adding transition layer elements and re-dividing tetrahedral elements. The final solid mesh is output for simulation analysis of resin transfer molding process flow and curing.

[0007] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method to address the core problems encountered in resin transfer molding processes for composite variable-angle layup parts. These problems include incorrect material region nesting and attribution due to complex surfaces and multiple variable-angle layups, overlapping or gaps between layup interfaces, and inconsistent topology of the final solid mesh. These issues hinder the generation of high-quality continuum meshes for flow and curing simulation analysis. The invention first constructs a local normal field based on the triangular mesh of the part surface. Starting from the bottom layup, it offsets layer by layer along the normal to generate an initial reference surface, enhanced by implicit distance field and thickness mapping. Then, it obtains preliminary and supplementary nesting and attribution labels through a combination of point inclusion testing and symbolic distance calculation. Next, it optimizes the material region partitioning using a region growth algorithm and adjacent consistency constraints. Using this as boundary constraint, it grows voxels layer by layer along the normal to form an initial volume mesh, achieving interlayer bonding through interface overlap and gap detection and local point displacement adjustment. Finally, it uses voxel repair and tetrahedral repartitioning to address remaining topological defects, ensuring a geometrically closed, materially continuous, non-overlapping, and gap-free final solid mesh. This invention significantly improves the quality and reliability of solid mesh generation for complex curved surface parts of composite materials with variable angle layup, providing an efficient and accurate digital model foundation for subsequent process simulation analysis. Attached Figure Description

[0008] Figure 1 This is a flowchart of an automatic RTM mesh generation method based on implicit distance mapping and normal growth according to the present invention.

[0009] Figure 2 This is a schematic diagram of the module architecture of an automatic RTM mesh generation method based on implicit distance mapping and normal growth according to the present invention.

[0010] Figure 3 This is a detailed flowchart illustrating the automatic generation method of RTM mesh based on implicit distance mapping and normal growth according to the present invention.

[0011] Figure 4 This is a visualization diagram of the implicit distance field and normal field fusion of the present invention.

[0012] Figure 5 This is a visual schematic diagram of the three-level grid repair process of the present invention.

[0013] Figure 6 This is a cross-sectional view of the composite material variable angle layup part of the present invention.

[0014] Figure 7 This is a schematic diagram of the RTM mold cavity injection molding device of the present invention.

[0015] Figure 8 This is a schematic diagram of the voxel unit and triangular mesh spatial structure of the present invention.

[0016] Figure 9 This is a curve comparing the error rates of nested space attribution tags in this invention.

[0017] Figure 10 This is a thermal distribution diagram illustrating the simulation accuracy of the RTM process in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0019] In this application, implicit distance mapping refers to the process of mapping discrete points on the geometric surface of a part to a continuous implicit distance field and calculating the symbolic distance value to achieve continuous expression of geometric features; local normal field refers to the set of vectors representing the orientation of each point on the surface of the part, formed by organizing the unit normal vectors of all triangular facets on the surface of the composite material part according to the facet index; nested space attribution label refers to an array of integer identifiers that correspond one-to-one with the triangular facet or voxel element index, and the integer value of the array uniquely corresponds to the layup level number (the bottom layer is marked as 1, the next layer is marked as 2, and so on), used to quickly index the material level to which the geometric element belongs; initial volume mesh refers to the layup mesh obtained by growing voxels along the local normal field with the optimized material region as the boundary, which is different from the initial solid mesh in the simulation stage (referring to the basic mesh generated based on the prefabricated geometric structure and used for RTM process simulation); transition layer element refers to a thin layer of three-dimensional mesh element that is adapted to the original voxel element and generated to repair the topologically inconsistent region, used to achieve continuous transition of the material region.

[0020] like Figure 1 As shown, the automatic RTM mesh generation method of the present invention includes the following steps: Step S101 is to obtain the triangular mesh of the composite material part surface, and to obtain the local normal field of the part surface by calculating the unit normal vector of each triangular facet; Step S102 is to offset along the normal direction starting from the bottommost ply for the variable angle ply sequence to obtain the initial reference surface of the ply; Step S103 is to use the implicit distance field as the geometric representation basis, and convert the ply thickness into a distance offset value according to the thickness mapping relationship to generate the enhanced initial reference surface; Steps S104 to S106 are double-layer nested attribution label determination and region growth correction; Steps S107 to S109 are voxel growth and three-level mesh repair; Step S110 is to output the final solid mesh for RTM process simulation.

[0021] like Figure 2As shown, the RTM mesh automatic generation system of this invention comprises three layers: an input layer, a core processing layer, and an output layer. The input layer includes three types of data sources: part surface triangular meshes, variable-angle layup sequence data, and a material parameter database. The core processing layer sequentially comprises six processing units: a local normal field construction module, an implicit distance field offset module, a nested attribution label determination module, a region growth correction module, a voxel growth module, and a mesh repair module. These modules are serially connected via data flow. The output layer outputs the final solid mesh for use in RTM process flow and solidification simulation analysis. The material parameter database interacts with the implicit distance field offset module through a thickness mapping relationship, providing conversion parameters from layup thickness to distance offset values.

[0022] Specifically, this embodiment of an automatic RTM mesh generation method based on implicit distance mapping and normal growth may include: S101. Obtain the triangular mesh on the surface of the composite material part, and obtain the local normal field of the part surface by calculating the unit normal vector of each triangular facet.

[0023] The original triangular mesh data is obtained by reading the digitized part file. For each triangular facet in the original triangular mesh data, its normal vector is calculated using the vector cross product method and then normalized to obtain the unit normal vector. The unit normal vectors of each triangular facet are organized according to the facet index order to obtain a complete local normal field data set. Based on the unit normal vectors between adjacent triangular facets in the local normal field data set, the angle between the normal vectors of adjacent facets is calculated. If the angle between the normal vectors of adjacent facets is greater than a preset threshold, it is determined that there are surface protrusions or depressions in the current region, and the corresponding triangular facets are marked. Using the marked triangular facets and their unit normal vector directions, a continuous set of surface feature regions is obtained by expanding using a region growing method. Based on the statistical distribution of the unit normal vectors in the continuous set of surface feature regions, the main orientation category of the part surface is determined, and the classification result is output.

[0024] In this step, the preset threshold for the angle between the normal vectors of adjacent facets is set to 60°; if the surface of the part is a complex curved surface with high curvature (such as aircraft blades or wind turbine blade tips), the threshold can be adjusted to 45°. Triangular facets that are determined to have surface protrusions or depressions must simultaneously satisfy the condition that the angle between the normal vectors is greater than the threshold and the deviation of the normal vector direction is greater than 20°.

[0025] Obtaining the raw triangular mesh data by reading the digitized part file is the starting point of the entire process. For example, after reading the STL file of a car engine block, a mesh structure composed of a large number of triangular faces can be directly obtained. These faces collectively describe the geometric information of the part's outer surface. Calculating the normal vector for each triangular face using the cross product method is a common practice.

[0026] Specifically, the cross product of any two side vectors on the triangular facet is performed, and the result is the direction of the normal vector of the facet. Then, the normalization is achieved by dividing by its magnitude, thus obtaining a unit normal vector with a length of 1.

[0027] In one possible implementation, for a triangular facet in a planar region of a part, its unit normal vectors may all point in the positive Z-axis direction, while in a cylindrical region they may be radially distributed. By organizing the unit normal vectors of all triangular facets in facet index order, a complete set of local normal field data can be formed, which essentially records the orientation information at every point on the part's surface.

[0028] It should be noted that, based on the unit normal vector between adjacent triangular facets, the angle between them can be further calculated. Usually, the cosine value is obtained using the vector dot product formula, and then the actual angle is obtained through the inverse cosine function.

[0029] For example, on the surface of a part with a prominent boss, the angle between the normal vectors of adjacent facets may be close to zero degrees, while at the transition point of the boss edge, the angle may suddenly increase to close to 90 degrees or even greater. If the angle between the normal vectors of adjacent facets is greater than a preset threshold, for example, a threshold of 60 degrees, then it is determined that there is a surface protrusion or depression feature in the current area and the corresponding triangular facet is marked.

[0030] In one embodiment, for aircraft wing skin parts, when the included angle of certain adjacent facets exceeds a threshold, these facets are marked as feature regions that may belong to reinforcing ribs or grooves. Using the marked triangular facets and their unit normal vector directions, a continuous set of surface feature regions is obtained through region growing. Specifically, starting with a seed facet, it checks whether its adjacent facets also satisfy the included angle condition and have similar normal vector directions; if so, they are included in the same region, until further expansion is impossible. Based on the statistical distribution of unit normal vectors in the continuous set of surface feature regions, the main orientation category of the part surface can be determined, and the classification result is output.

[0031] For example, after statistically analyzing all unit normal vectors within a certain region, if most vectors are found to point in the three main directions—positive X-axis, positive Y-axis, and positive Z-axis—then the surface of the part can be determined to be primarily composed of a family of planes along the X, Y, and Z directions. In a preferred implementation, for complex curved surfaces, cluster analysis can be used to further subdivide the orientation categories, thereby supporting subsequent applications such as machining direction optimization or assembly posture planning. This method can effectively identify the geometric feature distribution of a part's surface, providing a reliable basis for feature recognition and process decision-making in intelligent manufacturing.

[0032] S102. For a variable angle ply sequence, starting from the bottommost ply, based on the ply angle and ply thickness, offset along the normal direction on the basis of the local normal field to obtain the initial reference surface of the ply.

[0033] Obtain all ply angles and corresponding ply thicknesses in the variable-angle ply sequence. Based on the bottom ply angle and thickness, extract the corresponding unit normal vector from the local normal field of the part. Perform an offset calculation along the extracted unit normal vector direction, according to the ply thickness value, to obtain the initial reference surface of the bottom ply. Use the initial reference surface of the bottom ply as the base surface for the next ply, and update the current ply normal reference based on the next ply angle. If there is a difference between the current ply angle and the previous ply angle, redistribute the normal field distribution on the current base surface according to the angle difference to obtain the adjusted normal field. Offset again along the normal direction using the adjusted normal field and the current ply thickness to obtain the initial reference surface of the current ply. Use the current initial reference surface as the base surface for the next ply, and repeat the angle update, normal field adjustment, and thickness offset process until all ply sequences are completed.

[0034] For example, in the field of composite material manufacturing, variable angle layup sequence refers to the arrangement of the fiber orientation of each layer as the position changes during the composite material lamination process. This sequence is usually generated by design software to optimize the mechanical properties of the part.

[0035] Specifically, all ply angles and corresponding thickness data can be obtained from CAD models or dedicated ply design files. For example, the ply sequence of an aircraft wing spars component may include angles such as 0 degrees, 45 degrees, and 90 degrees, with each layer having a thickness of 0.2 millimeters. These data ensure the uniform distribution and strength balance of the material.

[0036] like Figure 6 As shown, this invention demonstrates the cross-sectional structure of a composite material variable-angle layup part. Eight layers of composite material are laid within the mold cavity boundary formed by the upper and lower molds, using a symmetrical layup sequence [0 / 45 / 90 / -45]s. In the figure, layups 1 and 8 are 0-degree angle layups with the fiber direction horizontal; layups 2 and 7 are 45-degree angle layups; layups 3 and 6 are 90-degree angle layups with the fiber direction vertical; and layups 4 and 5 are -45-degree angle layups. Each layup layer is 0.2 mm thick, with a total thickness of 7.2 mm. The dashed lines in the figure represent the plane of symmetry, and the layup sequence is mirror-image distributed about the plane of symmetry. Different gray levels represent differences in fiber orientation at different layup angles.

[0037] In one embodiment, when extracting the corresponding unit normal vector from the local normal field of the part based on the bottom layer's layup angle and thickness, a case of a carbon fiber reinforced composite plate can be considered. The bottom layer has a layup angle of 0 degrees and a thickness of 0.15 mm. The local normal field is a set of unit vectors obtained from previous mesh calculations. Based on this, offset calculations are performed along these vector directions to obtain an initial reference surface, for example, with an offset distance equal to the thickness value. The surface is then simulated using a point cloud adjustment algorithm to ensure that the surface reflects the actual contour of the underlying material.

[0038] For example, after using the initial reference surface of the bottommost ply as the base surface of the next ply, the normal reference of the current ply needs to be updated according to the angle of the next ply. Assuming that the angle of the next layer is 45 degrees, which is different from the 0 degrees of the previous layer, the normal field distribution is redistributed on the base surface to obtain the adjusted normal field.

[0039] Specifically, this adjustment can be achieved through interpolation methods, such as calculating new vector directions at curved mesh points based on the angle difference. For example, when the angle difference is 45 degrees, the normal vector is rotated to match the fiber orientation, thereby maintaining the continuity of the structure.

[0040] In one embodiment, the initial reference surface of the current ply is obtained by adjusting the normal field and the current ply thickness and shifting them again along the normal direction.

[0041] For example, in the design of a composite material for a car bumper, the current thickness is 0.25 mm. The offset process involves moving the vertices of the surface point by point to generate new surface layers, which helps to simulate the effect of multi-layer stacking.

[0042] For example, the initial reference surface of the current layup can be used as the base surface for the next layup, and the angle update, normal field adjustment, and thickness offset processes can be repeated until all layup sequences are completed. In a wind turbine blade manufacturing example, the entire sequence has 10 layers, and processing each layer one by one ensures the accuracy of the final surface. This iterative approach can bring precise thickness control and reduce material waste in business applications.

[0043] In one embodiment, the entire process is connected by the fact that the generation of each surface depends on the output of the previous layer, forming a logical chain.

[0044] For example, offsets starting from the bottom layer provide a continuous foundation for subsequent layers, avoiding gaps between layers.

[0045] S103. Using the implicit distance field as the basis for geometric representation, the ply thickness is converted into a distance offset value according to the thickness mapping relationship to generate the enhanced initial reference surface.

[0046] Obtain the implicit distance field data of the original geometry. Calculate the distance offset value corresponding to the thickness of each ply using thickness mapping rules. Perform a global offset operation on the initial implicit distance field using the distance offset values ​​to obtain the enhanced distance field. Extract the zero-level set from the enhanced distance field to generate the initial reference surface after thickness enhancement. Discretize the initial reference surface after thickness enhancement to obtain a discrete reference mesh. If self-intersecting regions exist in the discrete reference mesh, eliminate the self-intersections through local signed distance reconstruction to obtain a corrected reference mesh. Output the enhanced reference surface that can be used for subsequent ply design based on the corrected reference mesh.

[0047] For example, in the design process of composite material wind turbine blades, it is first necessary to obtain the implicit distance field data of the original geometry. This data is essentially a mathematical representation that can describe the signed distance from any point to the nearest point on the blade surface. Positive values ​​represent the external space, and negative values ​​represent the internal space, thus providing a continuous geometric basis for subsequent thickness adjustments.

[0048] Specifically, for a typical wind turbine blade model, its original geometry is generated by CAD software and then converted into an implicit distance field mesh. In this implementation, the designer inputs the principal curvature profile data of the blade, such as a 0.5-meter length of surface point set from the blade tip to the blade root, and then uses a voxelization algorithm to calculate the distance value of each voxel, ensuring that the field data covers the entire design boundary area, thereby forming a high-resolution 3D distance field to support the accurate calculation of thickness enhancement.

[0049] like Figure 3 As shown, the method of this invention is processed in three stages. Stage 1 is the geometry construction stage, which sequentially executes five steps: triangular mesh reading, normal vector calculation, normal field organization, implicit distance field generation, and thickness mapping offset. Stage 2 is the attribution determination stage, which obtains preliminary attribution labels through spatial point inclusion testing and supplementary attribution labels through symbolic distance calculation. After region growth correction, the two types of labels are used to obtain optimized material region partitioning. Stage 3 is the mesh generation stage, which sequentially executes voxel growth operation, overlap gap detection, displacement adjustment, voxel repair, and tetrahedral partitioning. Among them, overlap gap detection is used as a judgment node. If there is a problem, it triggers the subsequent displacement adjustment and voxel repair processes.

[0050] like Figure 4As shown, this invention demonstrates the fusion process of the implicit distance field and the local normal field. The innermost gray surface in the figure represents the original part surface, corresponding to the zero level set where the signed distance d=0. The three outwardly offset surfaces correspond to the reference surfaces of ply 1, ply 2, and ply 3, with offset distances of 0.3mm, 0.6mm, and 0.9mm, respectively. The black arrows represent the unit normal vectors at each point, constituting the local normal field of the part surface. The implicit distance field is offset equidistantly in the direction of the normal field, converting the ply thickness into distance offset values ​​through a thickness mapping relationship, generating the enhanced reference surface corresponding to each ply. The horizontal dashed lines in the figure represent equidistant lines at different heights, reflecting the continuous distribution characteristics of the signed distance values ​​in space.

[0051] In one embodiment, the distance offset value corresponding to the thickness of each ply is calculated by a thickness mapping rule, where the thickness mapping rule is a predefined linear or nonlinear function that converts the actual thickness value in the ply sequence into an offset in the distance field.

[0052] For example, for a thicker ply at the leaf root with a thickness of 0.8 mm, it is mapped by rules to a distance offset of 0.4 mm to account for the material compressibility factor.

[0053] Specifically, the rules are set based on empirical formulas, such as the offset value being equal to half the thickness multiplied by a safety factor of 1.2, to avoid excessive expansion. In actual business, designers will first examine the workload distribution of the blade and select mapping rules to match the stress requirements of different areas. For example, the high curvature area in the middle section of the blade adopts a gradient mapping to ensure that the offset value gradually increases from 0.3 mm to 0.6 mm, forming a smooth transition.

[0054] Specifically, the thickness mapping relationship can adopt a linear mapping function, that is, the distance offset value is set as the product of the physical thickness of the layup and a preset compression factor, where the compression factor is usually between 0.85 and 0.95 to simulate the actual thickness compression of the composite material during the molding process; or, for large-thickness layup areas, a nonlinear attenuation mapping is adopted, that is, as the thickness increases, the increment of the distance offset value gradually decreases, thereby avoiding geometric interference caused by linear extrapolation and ensuring that the generated reference surface is physically reasonable and geometrically continuous.

[0055] In this application, the thickness mapping relationship adopts a combination of linear mapping and nonlinear attenuation mapping. The specific rules are as follows: 1. Linear mapping: applicable to conventional areas with a ply thickness ≤ 0.5 mm. The distance offset value = physical ply thickness × compression factor, with the compression factor ranging from 0.85 to 0.95. For composite carbon fiber prepreg, the compression factor is 0.9, and for glass fiber prepreg, it is 0.85. 2. Nonlinear attenuation mapping: applicable to thick areas with a ply thickness > 0.5 mm. As the ply thickness increases, the increment of the distance offset value gradually decreases. For every 0.1 mm increase in thickness, the increment of the distance offset value decreases by 10% to avoid geometric interference. 3. The distance offset value of all areas must be ≥ 0.05 mm to ensure the geometric continuity of the reference surface.

[0056] It should be noted that the initial implicit distance field is offset by the distance offset value to obtain the enhanced distance field. This operation is similar to the global translation of the field. The effect of increasing thickness is simulated by adding or subtracting the offset value at each point.

[0057] For example, in the process of enhancing wind turbine blades, after the initial distance field is loaded, the system will traverse all voxel points, directly add offsets for external positive points, and adjust accordingly for internal negative points to maintain the consistency of the field. This enhanced distance field not only preserves the original geometric topology but also expands the surface thickness.

[0058] In one possible implementation, this step is accelerated through parallel computing, ensuring that the computation time for a blade model with millions of voxels is controlled within a few seconds, thereby supporting rapid iterative design.

[0059] Specifically, the zero-level set is extracted based on the enhanced distance field to generate the initial reference surface after thickness enhancement. The zero-level set refers to the isosurface with a distance value of zero, which is the enhanced surface boundary. The extraction is achieved through the Marching Cubes algorithm. For example, in blade design, after the enhanced field is imported, the algorithm scans the voxel mesh, identifies the edges that cross the zero value and connects them to generate triangular patches, forming a continuous surface. This surface directly reflects the geometry after thickness enhancement, laying the foundation for subsequent layup path planning.

[0060] For example, the initial reference surface after thickness enhancement is discretized to obtain a discrete reference mesh. This process converts the continuous surface into a finite element mesh. For example, the Delaunay triangulation method is used to generate nodes and elements to ensure that the mesh density reaches 10 nodes per square centimeter in the high curvature region of the blade, so as to capture subtle shape changes and facilitate the integration of finite element analysis.

[0061] In one embodiment, if the discrete reference mesh has self-intersecting regions, the self-intersecting regions are eliminated by local symbolic distance reconstruction to obtain a corrected reference mesh. Self-intersecting regions often originate from mesh folding caused by complex curvature, such as mesh line intersections in the sharp region of the blade tip. The self-intersecting regions are repaired by identifying the intersection points and recalculating the symbolic distance field in the local region. The specific process includes isolating the problematic voxels, reconstructing the distance values, and retriangulating to ensure that the corrected mesh has no topological defects. This method is particularly suitable for high-precision blades in business applications to ensure simulation accuracy.

[0062] In this step, the specific process of eliminating self-intersections through local symbolic distance reconstruction is as follows: First, identify the self-intersecting regions of the discrete reference mesh and isolate the problem voxels by voxelization scanning; Second, recalculate the symbolic distance values ​​for the isolated problem voxels to correct the continuity of the distance field; Third, re-triangulate the self-intersecting regions based on the corrected symbolic distance values ​​to generate a local mesh without self-intersections; Fourth, stitch and merge the corrected local mesh with the original discrete reference mesh to obtain the corrected reference mesh.

[0063] It should be noted that the modified reference mesh output can ultimately be used as an enhanced reference surface for subsequent layup design. This output is exported in STL or IGES format. For example, in a wind turbine blade project, the modified mesh, after being smoothed, serves as the input basis for layup fiber path optimization, thereby achieving a seamless connection from geometric enhancement to structural reinforcement and improving overall design efficiency.

[0064] S104. Perform a spatial point inclusion test on the initial reference surface and the triangular mesh on the part surface to determine the ply area to which the center point of each triangular facet in the triangular mesh on the part surface belongs, and obtain a preliminary nested spatial attribution label.

[0065] Acquire initial reference surface and part surface triangular mesh data. Determine the positional relationship between the center point of each triangular facet and the initial reference surface using a spatial point inclusion test, obtaining preliminary nested spatial attribution labels. Based on the preliminary nested spatial attribution labels, group the triangular facets according to the ply region division rules to obtain the facet set corresponding to each ply region. Extract the center point coordinate sequence from the facet set of each ply region, arrange the center point coordinates according to the region boundary order, obtaining the region boundary center point sequence. Perform a planar projection transformation on the region boundary center point sequence, projecting it to the nearest point position on the initial reference surface, obtaining the projected center point sequence. By comparing the projected center point sequence with the curvature distribution of the initial reference surface, determine the uniformity of triangular facet distribution within each ply region, obtaining a region distribution uniformity label. Based on the region distribution uniformity label and the preliminary nested spatial attribution labels, reassign the triangular facets in unevenly distributed ply regions, obtaining optimized nested spatial attribution labels.

[0066] It should be noted that the "Nested SpaceAttribution Label" described in this application is specifically represented in terms of data structure as an array of integer identifiers that correspond one-to-one with the index of a triangular facet or voxel unit. Each integer value in the array uniquely corresponds to a layer level number (e.g., the bottom layer is marked as 1, and the next layer is marked as 2). This label is used to quickly index the material level to which a geometric element belongs in computer memory and is the logical basis for subsequent region growing and Boolean operations.

[0067] After obtaining the initial reference surface and the triangular mesh data of the part surface, it is first necessary to clarify the source and characteristics of these two types of geometric data. The initial reference surface is usually derived from the zero-level set surface obtained from the previous thickness enhancement processing. It represents the baseline geometry of the composite material layup design, while the triangular mesh of the part surface is a set of triangular facets obtained by discretizing the actual product shape. There is often a certain spatial deviation between the two. For example...

[0068] In one embodiment, for the design of composite material skin for a certain type of aircraft wing, the initial reference surface is a smooth neutral surface generated by the SDF offset method, while the triangular mesh on the part surface comes from an STL format mesh exported from the CAD model using finite element preprocessing software, containing approximately 500,000 triangular faces. By determining the positional relationship between the center point of each triangular face and the initial reference surface through spatial point inclusion testing, preliminary nested spatial attribution labels can be obtained.

[0069] Specifically, spatial point inclusion tests are often performed using ray casting or symbolic distance field lookup methods.

[0070] For example, for a triangular patch on the upper surface of an airfoil, whose center point coordinates are known, the shortest distance from this point to the initial reference surface is calculated, and the sign is used for judgment. If the center point is located outside the reference surface, it is marked as "external"; otherwise, it is marked as "internal," thus forming a preliminary spatial classification. Based on the preliminary nested spatial classification labels, the triangular patches are grouped according to the ply region division rules to obtain the patch set corresponding to each ply region.

[0071] In one possible implementation, the ply area division rules can be based on the geometric boundaries of ply areas at different angles such as 0°, 45°, 90°, and -45° as defined in the design drawings.

[0072] For example, for the skin of a vertical tail fin, the triangular facets are divided into three regions along the direction from the leading edge to the trailing edge: a reinforcement region, a transition region, and a main load-bearing region. The facet sets for each region are stored independently. The center point coordinate sequence is extracted from the facet sets of each ply region, and the center point coordinates are arranged in order according to the region boundaries to obtain the region boundary center point sequence.

[0073] Specifically, the center points of triangular facets on the boundary can be recorded sequentially using boundary tracing algorithms or region contour extraction methods to form an ordered chain of points.

[0074] For example, in the cylindrical region of the fuselage section, the sequence of boundary center points will appear as an approximately closed ring. A planar projection transformation is performed on the sequence of boundary center points of the region, projecting it to the nearest point on the initial reference surface to obtain the projected sequence of center points.

[0075] It should be noted that the projection process typically selects the nearest point to maintain geometric consistency. For example, after projecting a boundary point of a reinforcement zone onto the neutral surface, the point's position will shift slightly inward, reflecting the effect of curvature. By comparing the sequence of center points after projection with the curvature distribution of the initial reference surface, the uniformity of the triangular patch distribution within each ply region is determined, thus obtaining a label for the uniformity of regional distribution.

[0076] For example, when the spacing between projected points in a certain area does not match the local radius of curvature, such as excessively large point spacing in high curvature areas, it is judged as an uneven distribution and labeled as "non-uniform". Based on the regional distribution uniformity label and the initial nested spatial assignment label, the triangular facets in the unevenly distributed ply area are reassigned to obtain optimized nested spatial assignment labels.

[0077] In one embodiment, for non-uniform regions caused by dramatic curvature changes, some patches can be reclassified from the "outside" to the "inside," thereby making the ply boundaries more consistent with the actual geometric characteristics and ultimately improving the rationality of subsequent ply path planning.

[0078] S105. Project the points on the surface of the part into the implicit distance field through the thickness mapping relationship, calculate the symbolic distance value of each point, determine the nested region affiliation, and obtain the supplementary nested space affiliation label.

[0079] Point data on the surface of a part is processed using thickness mapping data. A projection operation is used to map these points into an implicit distance field, and the signed distance value of each point is calculated to obtain preliminary distance distribution data. Based on this preliminary distribution data, the signed distance value of each point is classified. If the signed distance value is greater than a preset threshold, it is classified as a first-class nested region; if the signed distance value is less than or equal to the preset threshold, it is classified as a second-class nested region, thus determining the region classification of each point. Point data from the first-class nested regions are extracted from their region classifications, and their deviation values ​​from the implicit distance field are calculated using a batch processing method to obtain the deviation distribution set for the first-class regions. For the point data of the second-class nested regions, the difference between their signed distance values ​​and preset thresholds is calculated through point-by-point comparison to obtain the difference distribution set for the second-class regions. The deviation distribution set for the first-class regions and the difference distribution set for the second-class regions are then integrated using a unified standardization method to obtain a comprehensive distribution dataset. Based on the comprehensive distribution dataset, a secondary verification is performed on the regional classification of each point. If the point deviation in the comprehensive distribution data exceeds a preset range, its classification is adjusted to a transitional region, resulting in an adjusted nested regional classification result. Using the adjusted nested regional classification result, a final supplementary nested spatial classification label is generated and saved to a preset database using a data storage tool, thus confirming complete regional classification information.

[0080] In this step, the preset threshold for the symbol distance value is set to 0.5mm, and the preset range for the position deviation is ±1.0mm. A symbol distance value greater than 0.5mm is a first-class nested region, and a symbol distance value less than or equal to 0.5mm is a second-class nested region. When the position deviation exceeds ±1.0mm, it is adjusted to a transition region. This threshold is suitable for the conventional design range of composite material ply thickness of 0.1-1.0mm.

[0081] In one possible implementation, point data on the surface of a part is processed through thickness mapping data. First, it's necessary to understand that thickness mapping data refers to thickness distribution information generated based on the part's 3D model. For example, in aerospace component manufacturing, thickness mapping can represent the material thickness variations on an airfoil surface. This point data is then mapped to an implicit distance field using a projection operation. An implicit distance field is a mathematical representation used to describe the function of the nearest distance from a point to a surface, typically represented symbolically as an internal or external location.

[0082] Specifically, the projection operation involves projecting each point onto the distance field grid along the normal direction. The process of calculating the symbol distance value is to solve for the shortest path from the point to the field boundary. For example, for a point P, its symbol distance SDF(P) = min(d(P,Q)) * sign, where d is the Euclidean distance and sign represents the inside and outside symbols. This yields preliminary distance distribution data, which is helpful for subsequent region division.

[0083] For example, when processing the surface of automotive parts, assuming the point data comes from a laser-scanned mesh model, with a preset threshold of 0.5mm, the symbolic distance value of each point is classified according to the preliminary distance distribution data. If it is greater than 0.5mm, it is classified as a first-class nested region, representing a thicker layered region; if it is less than or equal to, it is classified as a second-class region, representing a thin layered region, thereby determining the region category of each point.

[0084] In one possible implementation, point data for the first type of nested regions are extracted from the region classification. A batch processing approach is used to calculate the deviation values ​​from the implicit distance field. For example, a parallel computing framework such as OpenMP can be used to simultaneously calculate the deviation for a batch of points, i.e., the actual distance minus the expected value of the field, to obtain the deviation distribution set for the first type of regions. This can reveal the non-uniformity within the region. For the point data of the second type of nested regions, the difference is calculated by comparing each point individually. For example, the difference between the distance value of each point and the threshold of 0.5 mm is compared one by one, such as |SDF(P) - 0.5|, thus obtaining the difference distribution set for the second type of regions.

[0085] In one possible implementation, after obtaining the two sets of distributions, a unified standardization method, such as z-score normalization, is used to map the bias and difference values ​​to the same scale. For example, (x - μ) / σ is applied to the bias set, where μ is the mean and σ is the standard deviation. Then, the data is integrated into a comprehensive distribution dataset, which ensures the consistency of the data and facilitates subsequent analysis.

[0086] For example, a secondary verification is performed based on the comprehensive distribution dataset. If the deviation of the point exceeds the preset range, such as ±1.0, it is adjusted to a transition region, and the adjusted nested region assignment result is obtained. This process improves the accuracy of classification and can reduce structural weaknesses caused by misassignment in composite material layup.

[0087] In one possible implementation, the final supplementary nested space attribution label is generated by adjusting the results and saved using a database tool such as SQLite. This determines the complete area information and supports subsequent manufacturing optimization.

[0088] S106. Based on the initial nested space attribution labels, the region growth algorithm is used to correct the incorrectly assigned regions. The nested space is re-divided through the consistency constraint of adjacent unit labels to obtain the optimized material region division result.

[0089] Obtain preliminary nested spatial attribution labels. A region growing algorithm is used to identify erroneous labeling regions from the preliminary nested spatial attribution labels, resulting in a set of erroneous labeling regions. For this set, growth is performed based on label consistency constraints between adjacent units. If the current unit has the same label as an already confirmed region, it is merged into that region; otherwise, growth stops, resulting in a corrected preliminary region division. All adjacent unit pairs are extracted from the corrected preliminary region division, and their labels are checked for consistency. If inconsistent, they are marked as boundaries to be re-divided, resulting in a set of boundaries to be re-divided. For this set of boundaries to be re-divided, label consistency constraints are used to redistribute the attribution of units on both sides of the boundary, resulting in an adjusted nested spatial division. Based on the adjusted nested spatial division, the attribution labels of all units are traversed to determine the complete boundary of each material region, resulting in an optimized material region division result. A connectivity check is performed on the optimized material region division result. If isolated small regions exist, their attribution is corrected based on the labels of adjacent main regions, resulting in the final material region division result.

[0090] In this step, the specific execution rules of the region growing algorithm are as follows: 1. Seed point selection: Select the triangular facet / voxel unit at the geometric center of the ply region as the initial seed point. The seed point must satisfy the requirement that its belonging label is clear and there is no conflict with the surrounding units; 2. Growth determination: Expand layer by layer to the adjacent units with the seed point as the center. If the adjacent units have the same belonging label as the seed point, they are included in the growth region. If the labels are inconsistent, the growth in this direction is stopped; 3. Growth termination condition: The expansion reaches the geometric boundary of the ply region, or the adjacent units are all units with inconsistent labels; 4. Error region identification: Units that exist in isolation and whose labels are different from their own by most of the surrounding units are determined to be error regions.

[0091] like Figure 9As shown, this invention demonstrates a comparison of nested space attribution label error rates under different surface complexity levels. The horizontal axis represents the surface complexity level, from left to right: simple surface (curvature less than 0.1), medium surface (curvature 0.1-0.5), complex surface (curvature 0.5-1.0), high curvature surface (curvature greater than 1.0), and variable angle region (angle greater than 30 degrees). The vertical axis represents the attribution label error rate. The solid dotted curve represents the traditional method, and the dashed dotted curve represents the method of this invention. The curve comparison shows that the error rates of the traditional method at each complexity level are 8%, 12%, 18%, 25%, and 22%, averaging approximately 15%; the error rates of the method of this invention are 2%, 3%, 4%, 5%, and 4.5%, all below the target threshold of 5%. The method of this invention achieves an error rate reduction of 60%-80% at all complexity levels.

[0092] In one possible implementation, the process of obtaining preliminary nested spatial attribution labels can be understood as performing initial classification of point cloud data on the surface of the part.

[0093] For example, in the design of aero-engine blades, the surface points of the blade are first collected using scanning equipment. Then, the signed distance of each point is calculated based on an implicit distance field. Points with distance values ​​greater than 0 are marked as outer regions, and points with distance values ​​less than 0 are marked as inner nested regions, thus forming a preliminary set of labels. This preliminary labeling helps in subsequent error identification and ensures a rough division of material distribution.

[0094] For example, when using a region growing algorithm to identify erroneous regions from an initial nested spatial attribution label, it can start from a seed point and expand, checking the label consistency of neighboring points. If a point's label does not match that of most of its surrounding points, it is included in the erroneous region set. In an automotive parts casting model, assuming the initial labels incorrectly mark some boundary points as internal regions, the algorithm will start from the confirmed external points and grow layer by layer to identify those isolated internal labeled points, forming an erroneous region set. The key to this step is selecting appropriate seed points and growth thresholds to avoid over-expansion.

[0095] In one possible implementation, the process of determining growth based on the label consistency constraints of adjacent units for a set of labeled error regions involves iterative checks.

[0096] For example, if the label of the current cell is the same as that of an already confirmed area, the cell is merged; otherwise, the process stops. In building structure simulation, for a nested space of a concrete beam, if the erroneous area is due to label flipping caused by noise, the system checks the majority label votes of adjacent cells. If the majority are areas with reinforced concrete, the current cell is merged to obtain a corrected preliminary area division, thereby smoothing out the influence of noise and improving the robustness of the division.

[0097] For example, from the revised initial region division, all adjacent unit pairs are extracted, and their labels are checked for consistency. If they are inconsistent, they are marked as boundaries to be redefined. In the design of electronic device housings, after extracting adjacent unit pairs, if it is found that the labels at some boundaries change from plastic to metal inconsistently, these pairs are marked as sets to be processed, which helps in the precise definition of boundaries.

[0098] In one possible implementation, for the set of boundaries to be repartitioned, label consistency constraints are used to redistribute the ownership of the units on both sides of the boundary.

[0099] For example, the adjusted nested space partitioning is obtained by reassigning weighted average label values. In medical implant models, if labels conflict on both sides of the boundary, the system will redistribute them according to the distance field gradient to ensure a smooth transition between bone and filling material.

[0100] For example, the process of determining the complete boundary of each material region by traversing the ownership labels of all units based on the adjusted nested space partitioning can be achieved by scanning unit by unit and connecting the boundary lines of the same labels. In the design of composite material plates, this traversal forms closed boundaries, resulting in the optimized material region partitioning. This step ensures the continuity of the regions.

[0101] In one possible implementation, a connectivity check is performed on the optimized material region division results, and if isolated small regions exist, their affiliation is corrected based on the labels of adjacent main regions.

[0102] For example, in semiconductor chip layout, after an isolated pixel is identified, it is assigned to a neighboring silicon-based region, resulting in the final material region partitioning. This correction avoids fragmentation, improves the overall integrity of the model, and can reduce material waste and increase structural strength in actual production.

[0103] S107. Using the optimized material region as boundary constraint conditions, perform voxel growth operations layer by layer along the local normal field of each layer to generate the initial volume mesh corresponding to each layup.

[0104] Obtain the optimized material region partitioning results. Extract the boundary constraints for each layer based on the material region partitioning results. Calculate the local normal field direction for each layer from the boundary constraints. Initialize the voxel growth starting point layer by layer for each local normal field direction. If there are ungrown voxels in the current layer, expand the voxel elements along the local normal field direction to obtain the growth result for the current layer. Determine if all voxel elements in the current layer have completed growth; if so, fix the initial volume mesh for the current layer and proceed to the next layer. By combining the fixed initial volume meshes of each layer, obtain the initial volume mesh set corresponding to the complete layup.

[0105] In this step, the specific parameters and execution rules for voxel growth are as follows: 1. Voxel unit size: determined according to the ply thickness, voxel size = ply thickness / 2, a typical ply thickness of 0.2mm corresponds to a voxel size of 0.1mm; 2. Growth start point: the center point of the triangular facet of the ply region boundary is selected as the initial starting point for voxel growth, and the growth start point of each ply is aligned with the boundary of the previous ply; 3. Growth step size: the growth step size along the local normal field direction is consistent with the voxel size; 4. Growth termination condition: the voxel extends to the geometric boundary of the ply region, or the voxel unit does not overlap with the adjacent ply voxel unit and the gap is ≤0.02mm; 5. Variable angle ply adaptation: when the ply angle changes by ≥30°, the local normal field is recalculated and the growth direction is adjusted to ensure that the voxel growth is consistent with the fiber orientation.

[0106] For example, in the manufacturing of aerospace composite components, after obtaining the optimized material region division results, the boundary constraints of each layer are extracted based on these results. These constraints cover the geometric features of the interlayer transition zone and the material property matching requirements. For the plylay design of wing skin, the coordinate point set and normal vector information of each layer boundary are extracted by analyzing the contour data of the region division, serving as the basis for subsequent calculations.

[0107] It should be noted that the process of calculating the local normal field direction of each layer from the boundary constraints relies on the differentiation operation of the boundary curve to obtain the local direction vector perpendicular to the boundary.

[0108] like Figure 8 As shown, this invention illustrates the physical morphology and interrelationship of voxel units and triangular meshes in space. The left side of the figure shows a 3x3x2 voxel mesh structure, where each voxel unit is cubic in shape. The voxel size is determined by the ply thickness, typically half the ply thickness. The upper right side of the figure shows a triangular mesh surface, composed of multiple triangular facets, each with an outward-pointing unit normal vector. Enlarged views of individual voxel units and individual triangular facets are located at the bottom of the figure, with arrows indicating the direction of the triangular facet's normal vector. The voxel growth operation involves expanding voxel units layer by layer along the normal field direction of the triangular mesh surface to form the initial volume mesh corresponding to each ply.

[0109] In one possible implementation, for carbon fiber reinforced plastic laminates, when the boundary has a curved shape, the local normal field direction is adjusted to point towards the infill region, ensuring that the growth path is consistent with the fiber orientation, thereby improving the structural stiffness.

[0110] In one possible implementation, the voxel growth starting point is initialized layer by layer for each layer's local normal field direction. This step provides directional guidance for subsequent expansion.

[0111] For example, if there are ungrown voxels in the current layer, the voxel elements are expanded along the local normal field direction to obtain the growth result of the current layer. In the curing process of thermosetting resin-based composites, this expansion simulates the behavior of resin wetting fibers, avoiding the generation of dry spot defects. It is determined that all voxel elements in the current layer have completed growth; if so, the initial volumetric mesh of the current layer is fixed and the process moves to the next layer. When the growth fronts of all elements reach the layer boundary, the mesh of that layer is locked for multi-layer stacking. By combining the initial volumetric meshes of each fixed layer, the initial volumetric mesh set corresponding to the complete layup is obtained. In practical applications, this combination method can provide an accurate initial geometric model for finite element simulations, improving prediction accuracy.

[0112] S108. Perform interface overlap and gap detection on the initial volume mesh between adjacent layers to determine whether there is cell overlap or gap. Adjust the local mesh point displacement to make the cells at the interface fit together, and obtain an intermediate volume mesh without overlap or gap.

[0113] Obtain the initial volume mesh between adjacent layers. Perform interface overlap and gap detection on the initial volume mesh to determine if there are overlapping or gaps between elements. If overlapping is detected, obtain the coordinates of the mesh nodes within the overlapping region. If a gap is detected, obtain the coordinates of the mesh nodes within the gap region. Calculate the required displacement vector for each node based on the mesh node coordinates of the overlapping and gap regions. Apply displacement vectors to the local mesh nodes using a finite element mesh adjustment method to obtain the adjusted node positions. Update the topology of all elements in the initial volume mesh using the adjusted node positions to obtain an intermediate volume mesh without overlap or gaps.

[0114] In this step, the finite element mesh adjustment method used is the Laplace smoothing method. The specific execution method is as follows: 1. Establish a local sub-model for the mesh nodes in the overlapping / gap region. The sub-model includes the target node and three layers of adjacent nodes. 2. Apply displacement constraints to the target node according to the calculated displacement vector. The adjacent nodes are free nodes. 3. Iterate to solve the node position. The number of iterations is set to 10-20. After each iteration, check the element Jacobian determinant to ensure that there is no negative determinant. 4. After the iteration converges, update the coordinates of the target node and the surrounding nodes to complete the local mesh adjustment.

[0115] After obtaining the initial volume mesh between adjacent layers, it is necessary to pay attention to their actual spatial correspondence.

[0116] Specifically, in multidirectional layup design of composite materials, adjacent layups often have different fiber orientation angles, such as 0° layups and 45° layups adjacent to each other. In this case, it is difficult for the element boundaries of the initial volume mesh of the two layers to align naturally. For example...

[0117] In one possible implementation, assuming an aircraft wing sparsity region uses a [45 / -45 / 0 / 90]s symmetrical ply sequence, the initial volume mesh between the second 45° ply and the third 0° ply is first extracted. These two meshes are generated based on the local normal field of their respective layers; the element sizes may be identical, but the element center points and boundary surfaces do not completely coincide. The system compares the six faces of adjacent elements one by one, calculating the minimum distance between faces and the projected overlap area.

[0118] In one embodiment, when cell overlap is detected, for example, the top face of a 0° ply hexahedral cell intrudes approximately 0.3 mm into the bottom face of a 45° ply, the coordinates of all mesh nodes involved in the interference within the overlapping region are extracted. These nodes typically include the eight vertices of the overlapping volume and some in-plane insertion points. After obtaining the coordinates, the displacement vector that each node needs to translate outward along the local normal direction is calculated. The magnitude of the vector is proportional to the intrusion depth, while considering the stiffness weights of adjacent elements to avoid over-adjustment.

[0119] It should be noted that if a void is detected, such as an unfilled area approximately 0.25 mm wide between adjacent layers, the node coordinates on the void boundary are also recorded. In this case, the displacement vector direction changes to inward contraction, aiming to bring the meshes on both sides closer to the void center. In the specific calculation, the geometric center of the void region is preferentially selected as the target reference point, and then the displacement is allocated inversely proportional to the distance of each node from the center to ensure boundary continuity.

[0120] Preferably, when using the finite element mesh adjustment method, the aforementioned displacement vector is applied as a boundary condition to the local sub-mesh. Taking a typical wind turbine blade main beam as an example, assuming that a certain 0° / 90° interface has a small amount of overlapping and gap mixed region, the system first establishes a local sub-model containing hundreds of elements. After applying displacement constraints, the new coordinates of the nodes are solved iteratively to ensure that the Jacobian determinant of the adjusted elements is still positive, thus avoiding mesh distortion.

[0121] Understandably, after the node positions are updated, the topological connections of all elements remain unchanged, with only the coordinates shifting. When the intermediate mesh is finally obtained, the element faces between adjacent plies achieve precise fit, with neither volume overlap nor tiny gaps, providing a high-quality geometric foundation for subsequent fiber path planning and performance analysis.

[0122] S109. For the topologically inconsistent regions that still exist in the intermediate mesh, a voxel repair algorithm is adopted. By adding transition layer units and re-dividing tetrahedral units, a final solid mesh with geometrically closed and material continuous structure is obtained.

[0123] Obtain the topological connectivity of all elements in the intermediate mesh. Perform voxelization transformation on the topological connectivity to obtain the corresponding voxel mesh structure. Identify the locations of topologically inconsistent regions using the voxel mesh structure. If a topologically inconsistent region is identified, generate transition layer elements around that region. Fill the empty voxel spaces based on the positional relationship between the transition layer elements and adjacent voxels. Perform repair processing on the filled voxel mesh using a voxel repair algorithm to obtain a repaired voxel mesh. Re-divide the repaired voxel mesh with tetrahedral elements to obtain a geometrically closed and material-continuous final solid mesh.

[0124] like Figure 5 As shown, this invention employs a three-stage mesh repair process to ensure the geometric closure and material continuity of the final solid mesh. Stage one involves interface overlap and gap detection, performing cell overlap volume and interface gap distance detection on the initial volume mesh between adjacent layers to identify problematic areas. Stage two involves local mesh point displacement adjustment, using the Laplace smoothing method to calculate displacement vectors and update node coordinates, ensuring the Jacobian determinant is greater than zero and displacement convergence, resulting in an intermediate volume mesh without overlap or gaps. Stage three involves voxel repair and re-meshing, sequentially performing morphological closure operations, surface smoothing, transition layer element generation, and tetrahedral element re-meshing, ultimately outputting a geometrically closed and materially continuous final solid mesh with a distortion rate less than or equal to 0.6 and a topological consistency rate of 100%.

[0125] In one possible implementation, the processing of the intermediate mesh first requires extracting the topological connections between all its internal cells. This involves analyzing the node sharing of each cell and the adjacency of edges and faces.

[0126] For example, in the design of aerospace composite structures, suppose there is a carbon fiber reinforced composite wing model. The mesh contains thousands of tetrahedral elements. By traversing the mesh data structure, it is possible to identify which elements share the same node or edge, thereby constructing an adjacency graph. This graph represents the connection paths between elements, ensuring the accuracy of subsequent transformations.

[0127] Specifically, voxelization is the process of mapping these topological connections to a voxel mesh structure. A voxel mesh is similar to a three-dimensional pixel mesh, where each voxel represents a small cube volume. The principle of this transformation is to discretize the continuous geometry, making the originally complex topology easier to handle.

[0128] In one embodiment, for the aforementioned wing model, voxelization can employ a uniform mesh generation method, dividing the entire mesh space into cubic voxels of 0.5 mm in size. Then, based on the topological connections of the elements, the voxels corresponding to connected elements are marked as occupied, and the connection direction is recorded. The resulting voxel mesh structure reflects the continuous distribution of the material, avoiding the distortion problem of traditional meshes on curved surfaces. This transformation also simplifies boundary condition handling in computation, as the regularity of voxels facilitates the application of convolution operations to detect anomalies.

[0129] For example, identifying regions of topological inconsistency using a voxel mesh structure typically involves scanning all voxels and checking whether the connections between adjacent voxels conform to expected consistency rules. For instance, if two adjacent voxels should be connected but are actually disconnected, or if the connection directions are mismatched, these are marked as inconsistent regions. In the curved sections of an airfoil model, if ply transitions cause unexpected gaps between voxels, this identification process can traverse the voxel layers layer by layer, using a neighborhood search algorithm to find these locations, thereby pinpointing the source of the problem.

[0130] In one possible implementation, once a region of topological inconsistency is identified, transition layer elements need to be generated around that region. These elements act as bridges, by inserting additional geometric entities to smoothly connect them. For example, in the root connection region of an airfoil, if an inconsistency is detected, thin tetrahedral elements, half the size of the original voxel, can be generated, and their shape adjusted according to the density of the surrounding voxels. This ensures that the transition layer integrates with the original structure, maintaining material continuity and improving the accuracy of stress distribution during simulation.

[0131] Specifically, the empty voxel space is filled according to the positional relationship between the transition layer unit and the adjacent voxels. This step is achieved by calculating the positional offset and inserting new voxels.

[0132] For example, during the filling process, the nearest neighbor interpolation method can be used to extend the boundaries of the transition layer elements to the gaps, filling voxels one by one until the space is completely closed. In an airfoil model, this prevents leakage paths in aerodynamic simulations. A voxel repair algorithm is then used to repair the filled voxel mesh, resulting in a repaired voxel mesh. This algorithm typically includes smoothing and optimization steps, and its principle is to iteratively adjust the voxel boundaries to eliminate noise.

[0133] As a preferred embodiment, the voxel repair algorithm specifically includes two stages: morphological processing and smoothing processing. First, a three-dimensional morphological closing operation is performed on the filled voxel mesh, that is, an expansion operation is first performed to connect the broken voxel chains, and then an erosion operation is performed to restore the original volume, thereby filling the tiny holes. Subsequently, the vertex coordinates of the voxel surface are iteratively adjusted using the Laplacian smoothing algorithm or the average curvature flow algorithm to eliminate the stepped jagged features introduced by the voxelization process, and finally a topologically closed and surface-smooth repaired voxel mesh is obtained.

[0134] In this step, the specific rules for the generation of transition layer units and voxel repair are as follows: 1. Transition layer unit parameters: The size is half that of the original voxel unit, the generation density is 5 units / mm², and the unit type is the same as the original voxel unit (tetrahedral unit); 2. Transition layer unit generation: Transition layer units are generated within a 3-layer voxel range around the topologically inconsistent region, and the unit orientation is aligned with the normal field of the surrounding voxel units; 3. Void voxel filling: The nearest neighbor interpolation method is used to extend the boundary of the transition layer units to the void region, filling the void voxels one by one until the space is completely closed; 4. Morphological processing of the voxel repair algorithm: Three-dimensional morphological closing operation is used, with dilation operation iterating twice, erosion operation iterating twice, and smoothing processing using the Laplacian smoothing algorithm with 15 iterations.

[0135] In one embodiment, the Marching Cubes algorithm can be applied to reconstruct the surface after the mesh is filled, and then isolated voxels can be removed to obtain a smooth repaired voxel mesh. This can result in higher mesh quality and reduce computational errors in finite element analysis.

[0136] For example, by repairing the voxel mesh and re-dividing it into tetrahedral elements, a geometrically closed and materially continuous solid mesh is obtained. This involves inversely converting the voxel structure into a continuous mesh, using the Delaunay triangulation method to divide the tetrahedrons, and ensuring that all elements are closed and gapless. In the wing model, this final mesh can be used for strength testing, realizing a complete workflow from coarse mesh to fine simulation.

[0137] S110. Output the final solid mesh for simulation analysis of resin transfer molding process flow and curing.

[0138] Acquire the geometric structure data of the preform and generate an initial solid mesh using a mesh generation algorithm. Perform a quality check on the initial solid mesh; if the mesh distortion rate exceeds a preset threshold, perform mesh optimization to obtain an optimized solid mesh. Obtain resin viscosity, permeability, and curing kinetic parameters from the material database and load them into the finite element model corresponding to the optimized solid mesh. Set flow boundary conditions and pressure inlet conditions based on the mold cavity boundary and injection port location. Solve the resin flow control equations using the finite volume method to calculate the resin filling front position and pressure distribution field at each time step. Input the filling state and temperature field of the current time step into the curing reaction model to calculate the degree of curing distribution and exothermic rate. Iteratively execute the flow field calculation and curing reaction field calculation until the resin completely fills the cavity and the degree of curing reaches the set completion value, outputting the final flow and curing results on the solid mesh.

[0139] In this step, Darcy's law is used as the resin flow control equation, which states that the resin flow velocity in a porous medium is directly proportional to the pressure gradient and inversely proportional to the resin viscosity, thus adapting to the porous medium characteristics of the composite material preform. The curing reaction model adopts an n-stage autocatalytic model, which is suitable for thermosetting resins (epoxy resin, polyester resin) commonly used in RTM processes. The finite volume method is discretized using an upwind difference scheme, with a time step of 0.1 s and a pressure convergence accuracy of 10^-3 Pa. The criterion for completely filling the cavity with resin is a cavity filling rate ≥99.9%, and the degree of curing completion is set to 95% (the conventional curing requirement of RTM process).

[0140] In this application, the preset threshold for mesh distortion rate is set to 0.6. This threshold is the conventional quality threshold for composite material RTM process simulation mesh. When the distortion rate is greater than 0.6, mesh optimization is performed. After optimization, it is necessary to ensure that the mesh distortion rate is ≤0.6 and that the determinant of the element Jacobian matrix is ​​positive.

[0141] like Figure 7 As shown, the RTM mold cavity injection molding apparatus of the present invention includes an upper mold, a lower mold, a resin tank, an injection port, an vent, a vacuum pump, a temperature sensor, and a heating system. The upper and lower molds form a sealed cavity space, in which the composite material preform is placed. The resin tank injects resin into the cavity through the injection port, and the resin flows horizontally from left to right, impregnating the fiber reinforcement material in the preform. The vent is connected to the vacuum pump to evacuate the cavity before injection and to expel excess gas during injection. The temperature sensor monitors the cavity temperature, and the heating system controls the curing temperature. Typical process parameters include an injection pressure of 0.1-0.5 MPa and a curing temperature of 80-180 degrees Celsius. Sealing rings are provided on both sides of the mold to ensure the cavity is sealed.

[0142] For example, obtain the geometric structure data of the prefabricated body and generate an initial solid mesh through a mesh generation algorithm.

[0143] Specifically, for the prefabricated body at the root of the wind turbine blade, geometric information is extracted from the STEP format file, and an initial solid mesh is generated using the leading edge propulsion method. The initial solid mesh undergoes a quality check; if the mesh distortion rate exceeds a preset threshold, mesh optimization is performed to obtain the optimized solid mesh.

[0144] For example, by evaluating the quality indicators of cell shape, the optimization process adjusts the mesh topology to improve the overall quality.

[0145] like Figure 10 As shown in the figure, this invention demonstrates the spatial distribution of accuracy errors after RTM process simulation based on the mesh generated by this method. The horizontal axis represents the mold length direction, and the vertical axis represents the mold width direction. The shade of gray indicates the magnitude of the simulation error; darker colors indicate larger errors. The thermal distribution shows that the error is lower in the mold center area (2%-4%), while the errors in the edge and corner areas are relatively higher but still controlled within 8%. Contour lines mark the boundaries of four error levels: 2%, 4%, 6%, and 8%. Statistical data shows that the average error across the entire region is approximately 4.5%, the maximum error is less than 8%, and the minimum error is approximately 1.5%. Since the overall simulation error is less than the 8% threshold, it meets the accuracy requirements of RTM process simulation and can effectively guide the adjustment of process parameters in actual production.

[0146] In one embodiment, resin viscosity, permeability, and curing kinetic parameters are obtained from a material database and loaded into the finite element model corresponding to the optimized solid mesh.

[0147] Specifically, a polyester resin is selected, and its viscosity value at the injection temperature, preform permeability data, and curing reaction constant are loaded into the model material properties. Flow boundary conditions and pressure inlet conditions are set according to the mold cavity boundary and injection port location.

[0148] For example, the wall is made impermeable at the boundary, and constant pressure is applied at the inlet to drive the resin flow.

[0149] Specifically, the finite volume method is used to solve the resin flow control equations, calculating the resin filling front position and pressure distribution field at each time step. This method discretizes the computational domain into a series of finite volumes, discretizes the conservation equations in integral form, incorporates the permeability effect of porous media, uses an upwind difference scheme to handle the flow convection terms, and couples the velocity field through a pressure correction algorithm, advancing step by step to determine the dynamic position of the filling front and the pressure distribution of the entire domain.

[0150] For example, the filling state and temperature field at the current time step can be input into the curing reaction model to calculate the degree of cure distribution and heat release rate. This input drives the model to calculate the local curing progress and associated heat release rate based on reaction kinetics.

[0151] In one embodiment, the flow field calculation and curing reaction field calculation are performed cyclically until the resin completely fills the cavity and the degree of curing reaches the set completion value, and the final flow and curing results on the solid mesh are output.

[0152] Specifically, this process captures the dynamic changes during resin injection, ultimately providing comprehensive simulation data to guide the selection of process parameters in actual production.

[0153] 1. Part parameters: Composite material part for wind turbine blade main beam. The surface triangular mesh is read from STL file. There are 500,000 triangular facets. The part surface is a complex curved surface with high curvature. The layup sequence is [0 / 45 / 90 / -45] s symmetrical layup, with a total of 8 layers. The thickness of each layup layer is 0.2mm. Carbon fiber prepreg is used. The compression factor is 0.9.

[0154] 2. Step execution parameters: The threshold for the angle between the normal vectors of adjacent facets is set to 45°, the threshold for the symbol distance value classification is 0.5mm, the point deviation range is ±1.0mm, the mesh distortion rate threshold is 0.6, the voxel size is 0.1mm, the transition layer unit size is 0.05mm, the resin is epoxy resin, and the curing degree completion value is 95%.

[0155] 3. Execution Results: The final solid mesh of 8 layers generated by this method has no overlap or gaps, a topological consistency rate of 100%, a mesh distortion rate of 0.55%, a nesting space attribution label error rate of 3%, and a mesh generation time of 40 minutes. Based on the RTM process simulation, the simulated resin filling time has an error of 6% compared with the actual value, and the simulation results of the curing degree distribution match the actual test results by 92%, which fully meets the simulation requirements of actual production.

[0156] The practical application effects of this method are as follows: 1. Mesh generation efficiency: Compared with traditional manual mesh generation, the automatic mesh generation efficiency of this method is improved by more than 30%, and the mesh generation time for complex curved surface composite material parts is shortened from several hours to within 1 hour; 2. Assignment label accuracy: The error rate of nested spatial assignment labels is reduced from 15% of the traditional method to below 5%, and the material region division accuracy reaches 0.01mm level; 3. Mesh quality: The generated final solid mesh has no overlap or gaps, the mesh distortion rate is ≤0.6%, and the topology consistency rate is 100%, meeting the geometric requirements of RTM process simulation; 4. Simulation accuracy: Based on the mesh generated by this method, the RTM process simulation results have an error range of less than 8% compared with the actual process, which can effectively guide the adjustment of process parameters in actual production; 5. Adaptability: This method can be adapted to composite material variable angle ply layup parts with ply angles of 0°-90° and ply thicknesses of 0.1-1.0mm, and is suitable for complex curved surface composite material components in the fields of aerospace, wind power, and automobiles.

[0157] In this invention, the preset threshold for the angle between the normal vectors of adjacent facets is set to 60°. This threshold is determined based on statistical analysis of the surface geometric features of composite material parts. By analyzing surface triangular mesh data of a large number of typical composite material parts (including aircraft blades, wind turbine blade tips, automotive parts, etc.), it was found that when the angle between the normal vectors of adjacent facets is greater than 60°, it usually indicates that there are obvious protrusions or depressions on the surface of the part (such as reinforcing ribs, grooves, and areas of abrupt curvature change). For complex curved surfaces with high curvature (such as aircraft blades and wind turbine blade tips), the threshold can be adjusted to 45° to more accurately identify surface features. In this case, the triangular facets judged as having surface protrusions or depressions must simultaneously satisfy the condition that the angle between the normal vectors is greater than the threshold and the deviation of the normal vector direction is greater than 20°, in order to eliminate local noise interference caused by the discretization of the triangular mesh.

[0158] The preset threshold for symbol distance value classification is set to 0.5 mm. This threshold is determined based on the conventional design range (0.1-1.0 mm) of composite material ply thickness. In actual composite material manufacturing, ply thickness is typically between 0.1-1.0 mm, and 0.5 mm, as an intermediate value, effectively distinguishes regions of different thicknesses. When the symbol distance value is greater than 0.5 mm, it indicates that the point is located in a thicker ply region; when it is less than or equal to 0.5 mm, it indicates that the point is located in a thinner ply region. This threshold has been verified through extensive experiments to ensure the accuracy of nested region classification.

[0159] The preset range for point deviation is ±1.0 mm. This range is determined based on the manufacturing tolerance and measurement accuracy of the composite material ply thickness. In actual manufacturing, the tolerance of the composite material ply thickness is usually controlled within the range of ±0.1 mm to ±0.2 mm, while the calculation accuracy of the sign distance value can usually reach ±0.05 mm. Therefore, the deviation range of ±1.0 mm takes into account both manufacturing tolerance and avoids misjudgment caused by calculation errors, ensuring the reliability of the area attribution determination.

[0160] The preset threshold for mesh distortion rate is set to 0.6. This threshold is determined based on the conventional requirements for mesh quality in composite material RTM process simulation. In RTM process simulation, a mesh distortion rate (i.e., the deviation of the element shape from the ideal shape) exceeding 0.6 will lead to distorted simulation results and affect the accuracy of process parameter optimization. Through extensive simulation testing and verification using actual production data, meshes with a distortion rate ≤ 0.6 can ensure the reliability of simulation results; therefore, it is used as the threshold for mesh quality judgment.

[0161] The specific execution steps of the region growing algorithm used in this invention are as follows: 1. Seed point selection: Select the triangular facet at the geometric center of the ply region as the initial seed point. The seed point must satisfy the requirement that its belonging label is clear and that it does not conflict with the surrounding units.

[0162] 2. Growth Determination: Starting from the seed point, expand layer by layer to adjacent units, checking whether the labels of adjacent units match the seed point label. If they match, include the unit in the growth region; otherwise, stop growth in that direction.

[0163] 3. Growth termination condition: extension to the geometric boundary of the layup region, or adjacent elements are all elements with inconsistent labels.

[0164] 4. Error region identification: A unit that exists in isolation and whose labels are mostly different from its own are identified as an error region.

[0165] In practical applications, the region growing algorithm employs breadth-first search (BFS) to ensure the stability and efficiency of the growth process. During execution, the algorithm first initializes the seed point set, then expands adjacent units layer by layer, while maintaining a queue to manage units to be processed. Each processed unit is marked as completed to avoid duplicate processing.

[0166] The specific execution steps of the voxel growth algorithm used in this invention are as follows: 1. Voxel unit size determination: The voxel size is determined based on the ply thickness. Voxel size = ply thickness / 2. For example, a typical ply thickness of 0.2 mm corresponds to a voxel size of 0.1 mm.

[0167] 2. Determining the growth starting point: Select the center point of the triangular facet of the boundary of the ply region as the initial starting point for voxel growth, and align the growth starting point of each ply with the boundary of the previous ply.

[0168] 3. Determining the growth direction: Calculate the local normal field direction of each layer from the boundary constraints, and use it as the expansion direction of voxel growth.

[0169] 4. Growth process: If there are ungrown voxels in the current layer, the voxel units are expanded along the local normal field direction, and the expansion step size is consistent with the voxel size each time.

[0170] 5. Growth termination conditions: The voxel extends to the geometric boundary of the layup region, or the voxel unit does not overlap with the adjacent layup voxel unit and the gap is ≤0.02mm.

[0171] In practical applications, the voxel growth algorithm employs a depth-first search (DFS) approach to ensure the continuity and consistency of voxel growth. The growth process involves traversing the boundaries of the ply region, growing inward from the boundary points, and gradually filling the entire ply region. For plyes with variable angles (ply angle changes ≥30°), the algorithm recalculates the local normal field and adjusts the growth direction to ensure that voxel growth is consistent with fiber orientation.

[0172] The specific execution steps of the voxel repair algorithm used in this invention are as follows: 1. Voxelization Conversion: The topological connectivity of the intermediate body mesh is converted into a voxel mesh structure. The voxel mesh adopts a uniform mesh generation method, and the voxel size is consistent with the voxel size used in the voxel growth algorithm.

[0173] 2. Identification of inconsistent regions in topology: Scan all voxels and check whether the connection between adjacent voxels conforms to the expected consistency rules. If two adjacent voxels should be connected but are actually disconnected, they are marked as inconsistent regions.

[0174] 3. Transition layer element generation: Transition layer elements are generated within a 3-voxel range around the topologically inconsistent region. The size of the transition layer element is half that of the original voxel element, the generation density is 5 elements / mm², and the element orientation is aligned with the normal field of the surrounding voxel elements.

[0175] 4. Void filling: Using the nearest neighbor interpolation method, the boundary of the transition layer unit is extended to the void region, and void voxels are filled one by one until the space is completely closed.

[0176] 5. Voxel repair processing: Perform 3D morphological closing operations on the filled voxel mesh (dilation operation iterates 2 times, erosion operation iterates 2 times), and then use the Laplacian smoothing algorithm for smoothing processing (15 iterations) to obtain the repaired voxel mesh.

[0177] 6. Tetrahedral element re-division: Tetrahedral elements are re-divided by repairing the voxel mesh to obtain a final solid mesh that is geometrically closed and material-continuous.

[0178] In practical applications, the voxel repair algorithm first fills tiny voids through morphological closing operations, and then eliminates the stepped jagged features introduced by the voxelization process through Laplacian smoothing. This algorithm can effectively handle complex topologically inconsistent regions, ensuring that the generated mesh is geometrically closed and the material is continuous.

[0179] An implicit distance field is a mathematical representation used to describe the signed distance from any point in space to a target geometric surface. Positive values ​​indicate that the point is outside the surface, negative values ​​indicate that the point is inside the surface, and zero values ​​indicate that the point is on the surface. In this invention, the implicit distance field serves as the basis for geometric representation, used to accurately describe the surface geometric features and thickness variations of composite material parts.

[0180] The local normal field refers to the set of vectors formed by organizing the unit normal vectors of all triangular facets on the surface of a composite part according to their facet index order. This set represents the orientation information of each point on the part's surface and is the basis for ply offsetting along the normal direction. By calculating the angle between the unit normal vectors of adjacent triangular facets, the protrusions or depressions on the surface can be identified.

[0181] Nested space attribution tags are arrays of integer identifiers that correspond one-to-one with the indices of triangular facets or voxel units. Each integer value in the array uniquely corresponds to a layup level number (e.g., the bottom layer is labeled 1, and the next layer is labeled 2). These tags are used to quickly index the material level to which a geometric element belongs in computer memory and form the logical basis for subsequent region growing and Boolean operations.

[0182] The initial volume mesh refers to the layered mesh grown along the local normal field voxels, with the optimized material region as the boundary. It is different from the initial solid mesh in the simulation stage (referring to the basic mesh generated based on the prefabricated geometry and used for RTM process simulation), and is an intermediate result used for subsequent interface adjustment and mesh repair in this invention.

[0183] Transition layer elements are thin-layer 3D mesh elements adapted to the original voxel elements, generated to repair regions with topological inconsistencies. The size of the transition layer element is half that of the original voxel element, with a generation density of 5 elements / mm², and the element orientation is aligned with the normal field of the surrounding voxel elements. The function of the transition layer element is to smoothly connect inconsistent regions, ensuring material continuity and geometric closure.

[0184] The compression factor is a coefficient used to convert the physical thickness of a ply into a distance offset value, typically ranging from 0.85 to 0.95. This coefficient takes into account the thickness compression effect caused by pressure during the molding process of the composite material. In practical applications, a value of 0.9 is used for carbon fiber prepregs and 0.85 for glass fiber prepregs. The determination of the compression factor is based on extensive experimental data and industry experience, ensuring the accuracy of the thickness mapping.

[0185] Thickness mapping rules refer to the functional relationship that converts the physical thickness of a layup into a distance offset value, including two methods: linear mapping and nonlinear attenuation mapping. • Linear mapping: Applicable to conventional areas with a ply thickness ≤ 0.5 mm. Distance offset value = physical ply thickness × compression factor.

[0186] • Nonlinear attenuation mapping: Applicable to thick regions with ply thickness > 0.5 mm, the increment of distance offset gradually decreases as the ply thickness increases. Specifically, for every 0.1 mm increase in thickness, the increment of distance offset decreases by 10%, avoiding geometric interference.

[0187] The determination of the thickness mapping rule is based on experimental observation and theoretical analysis of the thickness change during the composite material layup process, ensuring that the reference surface after thickness enhancement is physically reasonable and geometrically continuous.

[0188] Voxel size refers to the size of each voxel unit during voxel growth, determined by the ply thickness: voxel size = ply thickness / 2. For example, when the ply thickness is 0.2 mm, the voxel size is 0.1 mm. The selection of voxel size takes into account the uniformity of ply thickness and the requirements of mesh resolution, ensuring the accuracy and efficiency of voxel growth.

[0189] Mesh optimization refers to the process of improving the quality of an initial solid mesh. When the mesh distortion rate exceeds a preset threshold (0.6), the mesh quality is improved by adjusting the mesh topology (such as moving nodes or re-meshing cells). The optimized mesh must ensure that the mesh distortion rate is ≤0.6 and that the Jacobian determinant of each element is positive, thus ensuring the geometric continuity and numerical stability of the mesh.

[0190] Through extensive practical application verification, the method of the present invention has achieved significant effects in the following aspects: 1. Mesh generation efficiency: Compared with traditional manual mesh generation, the automatic mesh generation efficiency of this method is improved by more than 30%, and the mesh generation time for complex curved surface composite material parts is shortened from several hours to within 1 hour.

[0191] 2. Accuracy of attribution labels: The error rate of nested spatial attribution labels has been reduced from 15% in traditional methods to below 5%, and the accuracy of material area division has reached the level of 0.01mm.

[0192] 3. Mesh quality: The generated final solid mesh has no overlap or gaps, a mesh distortion rate ≤0.6, and a topology consistency rate of 100%, meeting the geometric requirements of RTM process simulation.

[0193] 4. Simulation accuracy: Based on the mesh generated by this method, the RTM process simulation results have an error range of less than 8% compared with the actual process, which can effectively guide the adjustment of process parameters in actual production.

[0194] The above supplementary content details the key algorithms, models, thresholds, and other technical aspects involved in this invention, ensuring that those skilled in the art can implement this invention based on the description in the specification, in accordance with Article 26, Paragraph 3 of the Patent Law.

[0195] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An automatic RTM mesh generation method based on implicit distance mapping and normal growth, characterized in that, The method includes: Obtain the triangular mesh on the surface of the composite material part, and calculate the unit normal vector of each triangular facet to obtain the local normal field of the part surface; For a variable angle ply sequence, starting from the bottommost ply, the initial reference surface of the ply is obtained by offsetting along the normal direction based on the local normal field according to the ply angle and ply thickness. Using an implicit distance field as the basis for geometric representation, the ply thickness is converted into a distance offset value according to the thickness mapping relationship to generate an enhanced initial reference surface; Perform a spatial point inclusion test on the initial reference surface and the triangular mesh on the part surface to determine the ply area to which the center point of each triangular facet in the triangular mesh on the part surface belongs, and obtain a preliminary nested spatial attribution label. By projecting the part surface points into the implicit distance field through the thickness mapping relationship, calculating the symbolic distance value of each point, determining the nested region affiliation, and obtaining the supplementary nested space affiliation label; Based on the initial nested space attribution labels, the region growing algorithm is used to correct the incorrectly assigned regions, and the nested space is re-divided through the consistency constraint of adjacent unit labels to obtain the optimized material region division result; Using the optimized material region as the boundary constraint, voxel growth operations are performed layer by layer along the local normal field of each layer to generate the initial volume mesh corresponding to each layup. Interface overlap and gap detection is performed on the initial volume mesh between adjacent layers to determine whether there are cell overlaps or gaps. The cells at the interface are made to fit together by adjusting the local mesh point displacement to obtain an intermediate volume mesh without overlap or gaps.

2. The method according to claim 1, characterized in that, After obtaining the intermediate body mesh with no overlap and no gaps, the following is also included: For the topologically inconsistent regions that still exist in the intermediate mesh, a voxel repair algorithm is used to obtain a geometrically closed and materially continuous final solid mesh by adding transition layer elements and re-dividing tetrahedral elements. The final solid mesh is output for simulation analysis of resin transfer molding process flow and curing.

3. The method according to claim 1, characterized in that, The process of obtaining the triangular mesh on the surface of the composite material part, and calculating the unit normal vector of each triangular facet to obtain the local normal field of the part surface, includes: Obtain the original triangular mesh data by reading the digitized part file; For each triangular facet in the original triangular mesh data, the normal vector is calculated using the vector cross product method and then normalized to obtain the unit normal vector; The unit normal vectors of each triangular facet are organized according to the facet index order to obtain a complete set of local normal field data; Calculate the angle between the normal vectors of adjacent triangular facets based on the unit normal vectors between adjacent triangular facets in the local normal field data set; If the angle between the normal vectors of adjacent facets is greater than a preset threshold, it is determined that there are surface protrusions or depressions in the current region and the corresponding triangular facets are marked. Using the labeled triangular facets and their unit normal directions, a continuous set of surface feature regions is obtained by expanding them using a region growing method. Based on the statistical distribution of unit normal vectors in a continuous set of surface feature regions, the main orientation category of the part surface is determined and the classification result is output.

4. The method according to claim 1, characterized in that, For the variable-angle ply sequence, starting from the bottommost ply, based on the ply angle and ply thickness, the initial reference surface of the ply is obtained by offsetting along the normal direction from the local normal field, including: Obtain all ply angles and corresponding ply thicknesses in a variable-angle ply sequence; Based on the bottom layer's ply angle and ply thickness, extract the corresponding unit normal vector from the part's local normal field; Along the extracted unit normal vector direction, perform an offset calculation based on the ply thickness value to obtain the initial reference surface of the bottom ply. The initial reference surface of the bottommost ply is used as the base surface of the next ply, and the normal reference of the current ply is updated according to the angle of the next ply. If there is a difference between the current ply angle and the previous ply angle, the normal field distribution is redistributed on the current base surface according to the angle difference to obtain the adjusted normal field. By adjusting the normal field and the current ply thickness and shifting them again along the normal direction, the initial reference surface of the current ply is obtained; Use the initial reference surface of the current ply as the base surface for the next ply, and repeat the angle update, normal field adjustment and thickness offset process until all ply sequences are completed.

5. The method according to claim 1, characterized in that, The process involves using an implicit distance field as the geometric representation basis, converting the ply thickness into a distance offset value according to a thickness mapping relationship, and generating an enhanced initial reference surface, including: Obtain the implicit distance field data of the original geometry; The distance offset value corresponding to the thickness of each layer is calculated using thickness mapping rules; The enhanced range field is obtained by performing an overall offset operation on the initial implicit range field using the range offset value; The zero-level set is extracted from the enhanced distance field to generate the initial reference surface after thickness enhancement; The initial reference surface after thickness enhancement is discretized to obtain a discrete reference mesh. If the discrete reference mesh has self-intersecting regions, the self-intersecting regions are eliminated by reconstructing the local symbolic distance to obtain the corrected reference mesh; The modified reference mesh outputs an enhanced reference surface that can ultimately be used for subsequent ply designs.

6. The method according to claim 1, characterized in that, The step of performing a spatial point inclusion test on the initial reference surface and the triangular mesh of the part surface to determine the ply area to which the center point of each triangular facet in the triangular mesh of the part surface belongs, and obtaining a preliminary nested spatial attribution label, includes: Obtain the initial reference surface and part surface triangular mesh data; By using spatial point inclusion tests to determine the positional relationship between the center point of each triangular facet and the initial reference surface, preliminary nested spatial attribution labels are obtained; For the initial nested space attribution label, the triangular facets are grouped according to the ply area division rules to obtain the facet set corresponding to each ply area; Extract the center point coordinate sequence from the set of patches in each ply region, and arrange the center point coordinates according to the region boundary order to obtain the region boundary center point sequence; Perform a planar projection transformation on the sequence of center points of the region boundary, projecting it to the nearest point on the initial reference surface to obtain the sequence of center points after projection; By comparing the sequence of center points after projection with the curvature distribution of the initial reference surface, the uniformity of the triangular patch distribution within each ply region is determined, and a label for the uniformity of regional distribution is obtained. Based on the regional distribution uniformity label and the initial nested spatial assignment label, the triangular facets in the unevenly distributed ply regions are reassigned to obtain the optimized nested spatial assignment label.

7. The method according to claim 1, characterized in that, The process of projecting part surface points onto an implicit distance field using the thickness mapping relationship, calculating the signed distance value of each point, determining the nested region affiliation, and obtaining supplementary nested space affiliation labels includes: By processing the point data on the surface of the part through thickness mapping data, the point data is mapped to the implicit distance field using projection operation, and the sign distance value of each point is calculated to obtain preliminary distance distribution data. Based on the preliminary distance distribution data, the symbolic distance value of each point is classified. If the symbolic distance value is greater than the preset threshold, it is classified as the first type of nested region. If the symbolic distance value is less than or equal to the preset threshold, it is classified as the second type of nested region, thus determining the region category of each point. Extract the location data of the first type of nested region from the region classification, calculate the deviation value between it and the implicit distance field using batch processing, and obtain the deviation distribution set of the first type of region; For the point data of the second type of nested region, the difference between its symbol distance value and the preset threshold is calculated by comparing point by point to obtain the difference distribution set of the second type of region; Obtain the deviation distribution set of the first type of region and the difference distribution set of the second type of region. Use a unified standardization processing method to integrate the two distribution sets to obtain a comprehensive distribution dataset. Based on the comprehensive distribution dataset, a secondary verification is performed on the regional classification of each point. If the deviation of the points in the comprehensive distribution data exceeds the preset range, the classification is adjusted to a transitional region, resulting in the adjusted nested regional classification result. Based on the adjusted nested region attribution results, the final supplementary nested space attribution label is generated and saved to the preset database using a data storage tool to determine the complete region attribution information.

Images