Dirty geometry restoration method for complex three-dimensional geometric model

Abstract

The invention belongs to the technical field of geometric pretreatment in computer aided engineering (CAE), and particularly relates to a dirty geometry restoration method for a complex three-dimensional geometric model, which comprises a dirty geometry detection and restoration module based on heuristic rules and geometric topology analysis. By means of a multi-model collaborative alignment and bonding mechanism, high-precision assembly between parts is achieved through geometric feature matching and adjacent domain self-adaptive adsorption; acquiring material attributes provided by the simulation platform, and performing accurate cutting and boundary reconstruction on an intersection domain based on an intersection domain intelligent segmentation algorithm of the material attributes to ensure correct mapping of boundary conditions of a physical field; a geometric Boolean processing engine iterative learning mechanism of a general merging algorithm is integrated, and a high-quality model for simulating a grid is generated through multi-round iterative optimization. According to the method, double challenges of internal dirty geometry of a single model and derivative dirty geometry of intersection / adjacent areas among multiple bodies are solved, and key physical attributes and topological continuity of electromagnetic signal links are ensured and are completely reserved during repair.

Description

Technical Field

[0001] This invention belongs to the field of geometric preprocessing technology in Computer-Aided Engineering (CAE), specifically involving a dirty geometry repair method for complex three-dimensional geometric models, which is particularly suitable for multi-physics simulation scenarios with stringent electromagnetic integrity requirements, such as electronic systems, radio frequency devices, power modules, and multi-chip packages. Background Technology

[0002] In modern CAE multiphysics electromagnetic simulation, the geometric complexity and physical coupling strength of the simulation model have increased significantly, posing unprecedented challenges to the topological consistency, geometric fidelity and physical semantic integrity of the geometric model in the preprocessing stage.

[0003] Geometric preprocessing serves as a crucial bridge connecting CAD design and CAE simulation. Its core task is to convert the original computer-aided design (CAD) model into clean, manifold, and topologically conflict-free geometric input suitable for simulation meshes and numerical solutions. However, 3D models in real-world engineering commonly suffer from numerous defects known as "dirty geometry." These defects are difficult to handle in single-model scenarios and become even more complex in multi-model collaborative simulations. Issues such as assembly deviations between different components, blurred material interfaces, and broken signal links often lead to Boolean operation failures, mesh distortion, and even solution errors.

[0004] In existing technologies, most mainstream geometric restoration tools adopt rule-based local restoration strategies, lacking comprehensive consideration of multi-body collaborative relationships and physical semantics. Their main shortcomings include restoration isolation due to lack of multi-body collaboration, signal distortion caused by missing physical semantics, insufficient robustness of Boolean operations, and lack of closed-loop feedback and adaptive capabilities.

[0005] In recent years, although some studies have attempted to introduce machine learning for defect prediction or parameter recommendation, they have mostly remained at the classification level and have not yet achieved a complete closed-loop control from "diagnosis → source tracing → feedback → reprocessing", and lack explicit modeling of key physical constraints such as electromagnetic link continuity.

[0006] Therefore, there is an urgent need for an intelligent repair framework that can integrate geometric topology analysis, multi-body collaborative alignment, material-sensing segmentation, and closed-loop iterative optimization to achieve highly automated and high-fidelity preprocessing of complex 3D models, and to ensure the physical authenticity of electromagnetic signal links and the credibility of simulation results. Summary of the Invention

[0007] This invention provides a method for repairing dirty geometry in complex 3D geometric models, aiming to solve problems such as isolated geometry repair, lack of physical semantics, high failure rate of Boolean operations, and lack of closed-loop feedback in existing technologies. By constructing an integrated, iterative, closed-loop driven, and material-aware repair architecture, it achieves fully automated conversion from the original multibody CAD model to clean simulation geometry, ensuring the topological integrity, geometric manifold, and continuity of key physical links of the model, and significantly improving the robustness of subsequent mesh generation and the accuracy of simulation solutions.

[0008] To achieve the objectives of this invention, the technical solution adopted is: a dirty geometry repair algorithm framework for complex three-dimensional geometric models, comprising:

[0009] Step 1: A dirty geometry detection and repair module based on heuristic rules and geometric topology analysis is used to identify and eliminate model topology problems, as well as minor features and local defects. Heuristic rules are used to guide the identification, classification, and repair strategy selection of dirty geometric defects in Brep models. Heuristic rules refer to a class of incomplete but highly efficient decision criteria that integrate geometric prior knowledge, engineering experience, and physical rationality constraints. They are used to guide the identification, classification, and repair strategy selection of common dirty geometric defects in complex 3D Brep models. These rules do not rely on rigorous mathematical proofs but are summarized based on high-frequency defect patterns in a large number of actual engineering cases, and have good robustness, interpretability, operability, and computational efficiency advantages.

[0010] Step 2: Multi-model collaborative alignment and bonding mechanism, through geometric feature matching and neighborhood adaptive adsorption, to achieve high-precision assembly between parts;

[0011] Step 3: Obtain the material properties provided by the simulation platform. Based on the material properties, the intelligent segmentation algorithm for the intersection domain is used to accurately cut and reconstruct the boundary of the intersection domain to ensure the correct mapping of the physical field boundary conditions. This algorithm is a geometric Boolean segmentation method guided by material semantics and aimed at the integrity of physical functions. It is used to handle the spatial intersection regions between different components in complex 3D models.

[0012] Step 4: Integrate the geometric Boolean processing engine with the General Fuse Algorithm (GFA) iterative learning mechanism to build an architecture with enhanced topological robustness and a closed loop of defect-aware feedback. Generate a high-quality model for simulation mesh through multiple rounds of iterative optimization. Perform defect diagnosis after Boolean execution and continue processing to build a closed-loop optimization process of "GFA-diagnosis-defect feedback-reprocessing" to continuously improve model cleanliness.

[0013] As an optimized solution of the present invention, step 1 specifically includes:

[0014] Step 1.1: Based on the geometric topology data of the Brep model, and using heuristic rules—a type of intelligent judgment criterion summarized from experience, engineering knowledge, and geometric physical laws—to guide the identification, classification, and repair strategy selection of various defects in the B-rep model, this step is a strategy mechanism that balances computational efficiency and robustness, including defect identification, defect classification, and repair strategy guidance. The following are some key defect identification steps:

[0015] Topological anomaly identification is used to identify topological defects that violate manifoldness, closure, or adjacency consistency, including but not limited to: non-current edge detection where an edge is shared by three or more surfaces, dangling edge detection, isolated point detection, and inconsistent cycle orientation detection.

[0016] Geometric anomaly identification is used to identify geometric defects that are detrimental to mesh generation and engineering processing, including but not limited to: identification of small faces with area S / boundary length less than the engineering threshold, identification of line-surface self-intersection, identification of overlap, identification of spline surface curvature abrupt change anomaly, and identification of plane point non-plane due to tolerance anomaly.

[0017] Then, a joint topology-geometric anomaly detection mechanism is used. This mechanism combines topological anomalies, geometric anomalies, and contextual information such as physical field ports, excitations, and materials to classify defect features, match defect feature processing strategy groups (there are multiple defect processing strategies), and generate dirty geometric defect data.

[0018] Step 1.2: Based on the dirty geometric feature data, and based on the principle of minimum intervention and the physical connectivity preservation criterion, select a relatively superior strategy for local model repair;

[0019] The principle of minimum intervention is to select the solution that causes the least change to the original geometry among multiple feasible repair paths. The principle of minimum intervention includes, but is not limited to, minimizing changes to the topology of points, lines and surfaces, minimizing changes to volume, and minimizing deviations in surface curvature. These rules are comprehensively adapted according to the geometric defects.

[0020] The physical connectivity preservation criterion requires ensuring the connectivity of adjacent conductors remains unchanged during model repair to prevent physical link interruptions caused by model repair.

[0021] As an optimized solution of the present invention, step 2 specifically includes the following steps:

[0022] Step 2.1: For the multiple input Brep models, extract candidate feature sets through local differential geometry analysis and topological neighborhood query;

[0023] Step 2.2: Construct an alignment matching body based on the geometric features of the candidate feature set. The geometric features include, but are not limited to: planar features, cylindrical features, conical features, spherical features, toroidal features, rotating surface features, and stretched surface features. Using the finite distance function, axis alignment, edge fitting, and normal consistency measure, construct a feature matching map G = (V, E, F), where: V nodes are feature regions, E edges represent edge relationships, and F surfaces represent potential relationships.

[0024] Step 2.3: Based on the feature matching graph, construct the corresponding point set, edge, direction and other geometric features, and solve for the minimum weighted alignment error to suppress the influence of anomalies and achieve noise-resistant alignment correction.

[0025] Step 2.4: Based on the material properties provided by the simulation platform, generate a conductor network connectivity graph. Utilize topological neighborhood domains to perform model edge bonding. By deeply fusing the model's material properties, embed conductor connectivity constraints. That is, based on the material combinations of the neighborhood domains, perform differentiated processing between conductors, between dielectrics, and between dielectrics. Use the conductor network connectivity graph to verify the consistency of conductor connectivity and ensure the continuity of electromagnetic signal paths in geometric alignment and bonding.

[0026] Since this invention is applied to electromagnetic simulation, the material properties of the model are the simulated material properties, which include, but are not limited to, conductivity, dielectric constant, and magnetic permeability.

[0027] Step 2.5: After geometric alignment and gluing, perform global link connectivity analysis and verification. If the link connectivity verification fails or a new topology anomaly occurs, recovery is performed through a rollback mechanism. Then, local reconstruction is performed by extracting the surface information of the local area to ensure link connectivity.

[0028] As an optimized solution of the present invention, local differential geometric analysis and topological neighborhood query refers to identifying the parametric surface type with clear geometric semantics and extracting its key geometric features by calculating its differential attributes, including normal vector, principal curvature, Gaussian curvature, mean curvature, parametric gradient, etc., within the topological spatial neighborhood of each face or edge. The topological neighborhood includes the topological relationship of the model itself and the proximity relationship between models.

[0029] As an optimized solution of the present invention, in step 3, based on the model completed in steps 1 and 2, the model is further intelligently segmented into intersecting regions according to material properties (there are corresponding material properties in the simulation field, such as dielectric, conductor, conductivity, dielectric constant, electromagnetic properties, attenuation coefficient, etc.), which is built on a material semantic-guided geometric-physical collaborative segmentation mechanism. The specific steps are as follows:

[0030] Step 3.1: Based on the material properties of the model, classify the model and construct a material semantic graph M = {(Ωi, Mi)}, where Ωi is the i-th model and Mi is the material feature of the i-th model. The material features are classified according to conductivity, dielectric constant, etc., for subsequent segmentation strategy decision-making.

[0031] Step 3.2: Employ a hierarchical bounding box (BVH) structure to improve spatial collision detection, and use the GFA Boolean mechanism for segmentation to extract the geometric intersection region I = Ωi ∩ Ωj, thus identifying the intersection domain model;

[0032] Step 3.3: Based on the material semantic graph M, determine the attribution of the intersecting regions;

[0033] Step 3.4: After determining the material attribution, use GFA Boolean to perform model fusion and segmentation to ensure model boundary consistency.

[0034] As an optimized solution of the present invention, in step 3.3, the material properties of the two models are Mi and Mj, respectively, and the intersection domain determination rule is as follows:

[0035] In specific regional scenarios, the critical conductor region has the highest priority and is identified as an indivisible domain. Intersecting domains are then preferentially assigned to the model of this region. The critical conductor region refers to a highly sensitive conductor subdomain that plays a dominant role in electromagnetic energy transmission, signal integrity assurance, or boundary condition anchoring in the context of multiphysics coupling simulation. The geometric integrity of this type of region directly determines the stability of the field solution, the effectiveness of port excitation, and the accuracy of impedance continuity modeling in the subsequent finite element / boundary element discretization process.

[0036] In the case of similar materials, the intersection region belongs to the model with smaller geometric feature scale, and the geometric feature scale is measured by volume, area, etc.

[0037] In the case of materials of different types, the intersection region is assigned to the model with priority given to material grade, based on the material semantic graph.

[0038] As an optimized solution of the present invention, in step 4, based on the integrated GFA geometric Boolean processing engine, an architecture with enhanced topological robustness and defect-aware feedback loop is constructed, and a high-quality model for simulation mesh is generated through multiple rounds of iterative optimization. The specific steps are as follows:

[0039] Step 4.1: Based on the segmented model obtained in Step 3, perform global Boolean operations using GFA, and perform boundary calculations using a Brep-based geometric kernel to generate non-intersecting and non-overlapping geometric data Ωg and historical mapping history.

[0040] Step 4.2: Based on the geometric data Ωg generated in 4.1, perform defect analysis and detect typical abnormal patterns; according to the history, trace the source from the pre-Boolean model and construct the defect structure source table Mf.

[0041] = {(p,t,s,c)}, where p is the defect location, t is the defect type, s is the defect level, and c is the context information;

[0042] Step 4.3: Based on the defect source table Mf from step 4.2, perform classification processing. After defect processing, execute step 4.1 again to perform GFA iteration.

[0043] Step 4.4: Based on the geometric data Ωg generated in step 4.1 and the history mapping, generate geometric data adapted to the simulation mesh according to the mesh requirements (specifically related to the mesh requirements of the simulation software), merge the historical records of all previous steps, and generate a global context; through the above series of steps to optimize the model, achieve dirty geometry closed-loop correction, effectively and significantly improve the quality of simulation mesh generation, and greatly reduce human intervention.

[0044] As an optimized solution of the present invention, in step 4.2, typical abnormal patterns are detected, including but not limited to: topological anomalies caused by Boolean anomalies, such as non-manifolds, degenerate surfaces, and lost surfaces; minute features caused by Boolean anomalies, such as narrow faces, small faces, and gaps; and geometric distortions caused by Boolean anomalies, such as point tolerance anomalies. For topological anomalies, the model is regenerated based on boundary information, and the model is simplified to reduce the complexity of edge interference.

[0045] For minor features and gaps, single-model adaptive repair and multi-model boundary enhancement alignment and bonding are performed.

[0046] For geometric distortion, the model is merged with the material model to reduce edge interference.

[0047] As an optimized solution of the present invention, in step 4.3, the defect tracing table Mf is classified, including but not limited to:

[0048] If the defect structure traceability table Mf is empty, it means there is no defect feedback, then proceed to the next step 4.4;

[0049] If the defect type in the defect structure traceability table Mf is topological anomaly, then the model is regenerated based on the boundary information. The original surfaces of the topological model are used to find intersections and recalculate the model boundary. Complex surfaces in the model, such as spline surfaces, are simplified to reduce edge interference complexity.

[0050] If the defect type in the defect structure traceability table Mf is a small feature or gap, then single-model adaptive repair and multi-model boundary enhancement alignment and bonding are performed, that is, the repair functions in steps 1 and 2 are performed.

[0051] If the defect type in the defect structure traceability table Mf is geometric distortion, then the model is merged with the same material model to reduce edge interference.

[0052] This invention has positive and beneficial effects: 1) This invention achieves systematic breakthroughs in geometric processing integrity, physical semantic fidelity, computational robustness, and automation level. It constructs a four-level progressive processing architecture of "local repair - global alignment and bonding - interference segmentation - Boolean fusion iteration", which breaks through the limitations of traditional methods that are limited to isolated repair of single models. It also introduces a material-aware-driven collaborative alignment and bonding mechanism, establishes an intersection domain attribution determination system based on material level and functional priority, ensures the topological continuity of electromagnetic signal links, and constructs a closed-loop iterative learning mechanism of "GFA-diagnosis-feedback-reprocessing", which achieves a substantial improvement in the success rate of Boolean operations. It solves the key bottleneck problem in CAE preprocessing and constructs an intelligent repair ecosystem that integrates geometry, topology, materials, and functional semantics. It significantly improves the preparation efficiency, mesh generation quality, and solution convergence of electromagnetic simulation systems, and has important scientific value, engineering significance, and industrialization prospects.

[0053] 2) This invention achieves automated conversion from the original CAD model to high-fidelity simulation geometry through an integrated and closed-loop driven approach, significantly improving geometric cleanliness, topological consistency and physical link continuity, and providing a reliable input basis for subsequent mesh generation and simulation solution.

[0054] 3) This invention addresses the dual challenges of dirty geometry within a single model and dirty geometry derived from intersecting / adjacent regions between multiple bodies, while ensuring the topological continuity of key physical properties and electromagnetic signal links is fully preserved during the repair process.

[0055] 4) This invention incorporates material properties during the repair process to ensure the continuity of the electromagnetic link. By topologically tracing and verifying the connectivity of the conductor path, it prevents the interruption of the metal link due to geometric simplification or stitching operations, thus ensuring the realism of the current path in the electromagnetic simulation. The final output is a high-fidelity, manifold-like geometric model free of topological conflicts, significantly improving the robustness, element quality, and solution accuracy of subsequent mesh generation.

[0056] 5) This invention effectively solves the problems of automation and reliability in the geometric preparation stage of multiphysics simulation preprocessing, and is especially suitable for the simulation of complex systems with strict requirements for electromagnetic integrity, and has important engineering application value. Attached Figure Description

[0057] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0058] Figure 1 This is a schematic diagram of the process of the present invention;

[0059] Figure 2 This is a schematic diagram of the multi-model collaborative alignment and bonding mechanism;

[0060] Figure 3 This is a schematic diagram of an intelligent segmentation algorithm for intersection domains based on material properties.

[0061] Figure 4 A schematic diagram of the iterative learning mechanism for the geometric Boolean processing engine integrating GFA;

[0062] Figure 5 This is an example diagram of an application on the Rainbow Studio platform. Detailed Implementation

[0063] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0064] like Figure 1 As shown, this invention discloses a framework for dirty geometry repair algorithms for complex 3D geometric models. The method includes the following steps:

[0065] Step 1: Dirty geometry detection and repair module based on heuristic rules and geometric topology analysis;

[0066] The embodiments of this invention are based on the full-wave electromagnetic simulation platform Rainbow Studio EM software. This platform constructs a complete electromagnetic solution process, forming an integrated technical link of "3D modeling → dirty geometry repair → mesh generation → algorithm solution," covering four core functional modules, realizing a full-link simulation platform from modeling to solution. This invention focuses on the "dirty geometry repair" stage, repairing dirty geometry in 3D modeling or externally imported models, such as... Figure 1 and Figure 5 As shown, geometric modeling is performed using the geometric modeling platform of Rainbow Studio EM, a full-wave electromagnetic simulation platform. The model and material information are output. At the same time, based on the OpenCASCADE geometric kernel, the BRep model data is loaded and parsed. Topology-geometric joint anomaly detection is performed, defect feature data is matched, and local model repair is performed based on the principle of minimum intervention and the physical connectivity preservation criterion. The result is a clean single-unit model with consistent topology and no local defects.

[0067] Step 2: Implement a multi-model collaborative alignment and bonding mechanism through geometric feature matching and adaptive adsorption of neighboring regions.

[0068] In embodiments of the present invention, such as Figure 2Based on the model and material data generated in step 1, candidate feature sets are extracted from multiple input Brep models through local differential geometric analysis and topological neighborhood query. Geometric features are extracted using OpenCASCADE to construct an alignment matching volume. These features include, but are not limited to, parametric surface features such as planar features, cylindrical surface features, conical surface features, spherical features, toroidal features, surfaces of revolution, and stretched surface features. Finite distance functions, axis alignment, edge fitting, and normal consistency measures are used to construct a feature matching graph G = (V, E, F), where nodes V represent feature regions, edges E represent edge relationships, and surfaces F represent potential relationships. Based on the matching graph, corresponding point sets, edges, directions, and other geometric features are constructed and solved to calculate the minimum weighted alignment error, thus achieving model alignment.

[0069] Based on the material properties provided by the Rainbow Studio EM simulation platform, a conductor connectivity graph is generated. Using topological neighborhood domains, model edge bonding is performed. By deeply fusing the simulated material properties and embedding conductor connectivity constraints, the continuity of electromagnetic signal paths in geometric alignment and bonding is ensured.

[0070] After geometric alignment and gluing, a global analysis and verification of link connectivity is performed. The link connectivity is ensured through rollback mechanisms and local reconstruction.

[0071] Step 3: Intelligent segmentation algorithm for intersection domain based on simulated material properties;

[0072] In embodiments of the present invention, such as Figure 3 Based on steps 1 and 2, material properties provided by the simulation platform Rainbow Studio EM are obtained, models are classified, and a material semantic graph M = {(Ωi, Mi)} is constructed, where Ωi is the i-th model and Mi is the material feature of the i-th model. The material features are classified according to conductivity, dielectric constant, etc., for subsequent segmentation strategy decisions. A hierarchical bounding box structure (BVH) is adopted to improve spatial collision detection, and OpenCASCADE's GFA Boolean mechanism is used for segmentation to extract the geometric intersection region I = Ωi ∩ Ωj, which identifies the intersection domain model.

[0073] Based on the material semantic graph M, the intersection domain is determined. For two models with material properties Mi and Mj, the intersection domain determination rules are as follows:

[0074] In specific regional scenarios, the critical regions of the conductor have the highest priority and are identified as indivisible regions. In such cases, the intersecting regions are preferentially assigned to the model of that region.

[0075] In the case of the same material, the intersection region belongs to the model with a smaller geometric feature scale;

[0076] For different material cases, the intersection domains are assigned to the model with priority given to material grade, based on the material semantic graph.

[0077] After determining the material attribution, OpenCASCADE's GFA Boolean algorithm is used for model fusion and segmentation to ensure model boundary consistency and generate a model free of internal interference and material ambiguity.

[0078] Step 4: Integrate the geometric Boolean processing engine of GFA (General Fuse Algorithm) for iterative learning mechanism;

[0079] In embodiments of the present invention, such as Figure 4 Based on the model obtained in step 3, and according to the segmented model obtained in step 3, global Boolean operations are performed using OpenCASCADE's GFA to perform boundary calculations, generating non-intersecting and non-overlapping geometric data Ωg and a history mapping. Defect analysis is performed on the generated geometric data Ωg to detect typical abnormal patterns, including but not limited to: topological anomalies caused by Boolean, minor features caused by Boolean, and geometric distortions caused by other factors. Based on the history, the source is traced back to the pre-Boolean model, and a defect structure source table Mf = {(p,t,s,c)} is constructed, where p is the defect location, t is the defect type, s is the defect level, and c is the context information.

[0080] Based on the defect source table Mf, classification processing is performed, including but not limited to: if there is no defect feedback, proceed to the next step; if there is a topological anomaly, regenerate the model based on the boundary information, simplify the model, and reduce the complexity of edge interference; for minor features and gaps, perform single-model adaptive repair and multi-model boundary enhancement alignment and bonding; for geometric distortion, perform merging of models with the same material to reduce edge interference; after the above defect processing, GFA iteration is performed again.

[0081] Finally, geometric data adapted to the mesh generation, as well as model context information, are generated;

[0082] In the embodiments of this invention, based on the geometric modeling architecture of the full-wave electromagnetic simulation platform Rainbow Studio EM, this invention deeply integrates and reconstructs the underlying topological data structure and Boolean operation capabilities of the OpenCASCADE geometric kernel. This enables key modules such as dirty geometry detection and repair, multi-body alignment and bonding, material-aware segmentation, and GFA fusion iteration, constructing a high-fidelity geometric preprocessing system for complex 3D models. Specifically, relying on the physical field context provided by Rainbow Studio EM, a four-level progressive processing architecture of "local repair—global alignment and bonding—interference segmentation—Boolean fusion iteration" is constructed. A material-aware driven collaborative alignment and bonding mechanism is introduced, and an intersection domain attribution determination system based on material level and functional priority is established to ensure the topological continuity of the electromagnetic signal link. A closed-loop iterative learning mechanism of "GFA-diagnosis-feedback-reprocessing" is constructed. In implementation, this not only effectively repairs dirty geometric models but also significantly improves simulation fidelity and physical consistency. In summary, this invention not only provides an automated dirty geometry repair solution for complex 3D models, but also constructs an intelligent pre-processing ecosystem that integrates geometric topology analysis, material semantic understanding, and physical functional constraints. Its technical architecture has good scalability and engineering adaptability.

[0083] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for repairing dirty geometry in complex three-dimensional geometric models, characterized in that, include: Step 1: A dirty geometry detection and repair module based on heuristic rules and geometric topology analysis is used to identify and eliminate model topology problems, as well as small features and local defects; heuristic rules are used to guide the identification, classification and repair strategy selection of dirty geometry defects in the Brep model. Step 2: Multi-model collaborative alignment and bonding mechanism, through geometric feature matching and neighborhood adaptive adsorption, to achieve high-precision assembly between parts; Step 3: Obtain the material properties provided by the simulation platform, and use the intelligent segmentation algorithm of the intersection domain based on the material properties to accurately cut and reconstruct the boundary of the intersection domain, so as to ensure the correct mapping of the physical field boundary conditions; Step 4: Integrate the geometric Boolean processing engine with a general merging algorithm and an iterative learning mechanism to build an architecture with enhanced topological robustness and a defect-aware feedback loop. Generate a high-quality model for simulation mesh through multiple rounds of iterative optimization.

2. The method for repairing dirty geometry in complex three-dimensional geometric models according to claim 1, characterized in that, Step 1 specifically includes: Step 1.1: Based on the geometric topology data of the Brep model, including: Topological anomaly identification is used to identify topological defects that violate manifoldness, closure, or adjacency consistency, including but not limited to: non-current edge detection where an edge is shared by three or more surfaces, dangling edge detection, isolated point detection, and inconsistent cycle orientation detection. Geometric anomaly identification is used to identify geometric defects that are detrimental to mesh generation and engineering processing, including but not limited to: identification of small faces with area S / boundary length less than the engineering threshold, identification of line-surface self-intersection, identification of overlap, identification of spline surface curvature abrupt change anomaly, and identification of plane point non-plane due to tolerance anomaly. Then, a joint topology-geometric anomaly detection mechanism is used. This mechanism combines topological anomalies, geometric anomalies, as well as physical field ports, excitations, and materials to classify defect features, match defect feature processing strategy groups, and generate dirty geometric defect data. Step 1.2: Based on the dirty geometric feature data, and adhering to the principles of minimum intervention and physical connectivity preservation, select a strategy for local model repair; The principle of minimum intervention is to select the solution that causes the least change to the original geometry among multiple feasible repair paths. The principle of minimum intervention includes, but is not limited to, minimizing changes to the topology of points, lines and surfaces, minimizing changes to volume, and minimizing deviations in surface curvature. These rules are comprehensively adapted according to the geometric defects. The physical connectivity preservation criterion requires ensuring the connectivity of adjacent conductors remains unchanged during model repair to prevent physical link interruptions caused by model repair.

3. The method for repairing dirty geometry in complex three-dimensional geometric models according to claim 1, characterized in that: Step 2 specifically includes the following steps: Step 2.1: For the multiple input Brep models, extract candidate feature sets through local differential geometry analysis and topological neighborhood query; Step 2.2: Construct an alignment matching body based on the geometric features of the candidate feature set. The geometric features include, but are not limited to: planar features, cylindrical surface features, conical surface features, spherical features, toroidal features, surface of revolution features, and stretched surface features. Using the finite distance function, axis alignment, edge fitting, and normal consistency measure, construct a feature matching map G = (V, E, F), where: V nodes are feature regions, E edges represent edge relationships, and F surfaces represent potential relationships. Step 2.3: Based on the feature matching graph, construct geometric features including corresponding point sets, edges, and directions, and solve for the minimum weighted alignment error to suppress the influence of anomalies and achieve noise-resistant alignment correction. Step 2.4: Based on the material properties provided by the simulation platform, generate a conductor network connectivity graph. Utilize topological neighborhood domains to perform model edge bonding. By deeply fusing model material properties, embed conductor connectivity constraints. Based on the material combinations of the neighborhood domains, differentiate between conductors, between dielectrics, and between dielectrics. Use the conductor network connectivity graph to verify conductor connectivity consistency and ensure the continuity of electromagnetic signal paths in geometric alignment and bonding. Step 2.5: After geometric alignment and gluing, perform global link connectivity analysis and verification. If the link connectivity verification fails or a new topology anomaly occurs, recovery is performed through a rollback mechanism. Then, local reconstruction is performed by extracting the surface information of the local area to ensure link connectivity.

4. The method for repairing dirty geometry in complex three-dimensional geometric models according to claim 3, characterized in that: Local differential geometric analysis and topological neighborhood query refers to identifying parametric surface types with clear geometric semantics and extracting their key geometric features within the topological neighborhood of each face or edge by calculating its differential attributes, including normal vector, principal curvature, Gaussian curvature, mean curvature, and parametric gradient. The topological neighborhood includes the topological relationships of the model itself and the proximity relationships between models.

5. The method for repairing dirty geometry in complex three-dimensional geometric models according to claim 1, characterized in that: In step 3, based on the model completed in steps 1 and 2, the model is further intelligently segmented into intersecting domains according to material properties, building upon a material semantic-guided geometric-physical collaborative segmentation mechanism. The specific steps are as follows: Step 3.1: Based on the material properties of the model, classify the model and construct a material semantic graph M = {(Ωi, Mi)}, where Ωi is the i-th model and Mi is the material feature of the i-th model. The material features are classified into levels based on conductivity and dielectric constant. Step 3.2: Adopt a hierarchical bounding box structure to improve spatial collision detection, and use the GFA Boolean mechanism for segmentation to extract the geometric intersection region I = Ωi∩Ωj and identify the intersection domain model; Step 3.3: Based on the material semantic graph M, determine the attribution of the intersecting regions; Step 3.4: After determining the material attribution, use GFA Boolean to perform model fusion and segmentation to ensure model boundary consistency.

6. The method for repairing dirty geometry in complex three-dimensional geometric models according to claim 5, characterized in that: In step 3.3, the material properties of the two models are Mi and Mj, respectively, and the intersection domain determination rule is as follows: In specific regional scenarios, the critical conductor region has the highest priority and is identified as an indivisible domain. Intersecting domains are then preferentially assigned to the model of this region. The critical conductor region refers to a highly sensitive conductor subdomain that plays a dominant role in electromagnetic energy transmission, signal integrity assurance, or boundary condition anchoring in the context of multiphysics coupling simulation. The geometric integrity of this type of region directly determines the stability of the field solution, the effectiveness of port excitation, and the accuracy of impedance continuity modeling in the subsequent finite element / boundary element discretization process. In the case of similar materials, the intersection region belongs to the model with smaller geometric feature scale, and the geometric feature scale is measured using volume and area; In the case of materials of different types, the intersection region is assigned to the model with priority given to material grade, based on the material semantic graph.

7. The method for repairing dirty geometry in complex three-dimensional geometric models according to claim 1, characterized in that: In step 4, based on the integrated GFA geometric Boolean processing engine, an architecture with enhanced topological robustness and a defect-aware feedback loop is constructed. A high-quality model for simulation mesh is generated through multiple rounds of iterative optimization. The specific steps are as follows: Step 4.1: Based on the segmented model obtained in Step 3, perform global Boolean operations using GFA, and perform boundary calculations using a Brep-based geometric kernel to generate non-intersecting and non-overlapping geometric data Ωg and historical mapping history. Step 4.2: Based on the geometric data Ωg generated in 4.1, perform defect analysis and detect typical abnormal patterns; according to the history, trace the source to the pre-Boolean model and construct the defect structure source table Mf={(p,t,s,c)}, where p is the defect location, t is the defect type, s is the defect level, and c is the context information. Step 4.3: Based on the defect source table Mf from step 4.2, perform classification processing. After defect processing, execute step 4.1 again to perform GFA iteration. Step 4.4: Based on the geometric data Ωg generated in step 4.1 and the history mapping, generate geometric data adapted to the simulation mesh according to the mesh requirements, merge the historical records of all previous steps, and generate a global context; through the above series of steps, optimize the model to achieve dirty geometry closed-loop correction, effectively and significantly improve the quality of simulation mesh generation.

8. The method for repairing dirty geometry in complex three-dimensional geometric models according to claim 7, characterized in that: In step 4.2, typical anomalous patterns are detected, including but not limited to: topological anomalies caused by Boolean; minor features caused by Boolean and the resulting geometric distortions.

9. A method for repairing dirty geometry in complex three-dimensional geometric models according to claim 7, characterized in that: In step 4.3, the defect source table Mf is classified, including but not limited to: If the defect structure traceability table Mf is empty, it means there is no defect feedback, then proceed to the next step 4.4; If the defect type in the defect structure traceability table Mf is topological anomaly, then the model is regenerated based on the boundary information, the original surfaces of the topological model are used to perform intersection calculations to recalculate the model boundary, and the complex surfaces in the model are simplified to reduce the complexity of edge interference. If the defect type in the defect structure traceability table Mf is a small feature or gap, then single-model adaptive repair and multi-model boundary enhancement alignment and bonding are performed, that is, the repair functions in steps 1 and 2 are performed. If the defect type in the defect structure traceability table Mf is geometric distortion, then the model is merged with the same material model to reduce edge interference.

Images