Method and device for generating three-dimensional parameterized model based on artificial intelligence

Abstract

The invention discloses a three-dimensional parameterized model generation method and device based on artificial intelligence. The method comprises the following steps: receiving natural language description information which is input by a user and is used for describing a design demand of a mechanical part, and carrying out semantic analysis on the natural language description information to obtain structured demand data; based on the structured demand data, generating an initial form of the mechanical part by using a depth generation model, and performing geometric construction on the initial form to obtain a basic geometry of the mechanical part; and carrying out structure optimization and manufacturing feasibility verification on the basic geometry to generate a three-dimensional parameterized model meeting design requirements. The technical problem that the engineering precision is not high when an existing three-dimensional modeling technology is applied to mechanical parts is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular, to a method and device for generating a three-dimensional parametric model based on artificial intelligence. Background Art

[0002] Currently, the three-dimensional modeling technology of automotive manufacturing tooling equipment mainly relies on the manual design mode of traditional CAD software. Designers need to manually construct fixture structures such as locating pins, clamping arms, and support blocks, and repeatedly adjust geometric dimensions and assembly relationships. Although such methods can meet the design requirements of complex fixtures, their efficiency is limited by the experience of engineers and the degree of automation is low. To further improve efficiency, parametric modeling technology has been introduced into the generation process of standard tooling parts (such as cylinder mounting seats, guide blocks), and size adjustment is achieved through predefined templates and parameter input. However, this method is difficult to adapt to the design requirements of non-standard fixture structures such as special-shaped support arms and multi-degree-of-freedom adjustment mechanisms, and still relies on manual intervention. In recent years, 3D shape generation methods based on deep learning (such as VAE, GAN, diffusion models) have begun to be applied to the automated generation of historical fixture model data. However, due to limitations in model generalization ability and lack of physical constraints, problems such as topological structure errors and assembly interference are likely to occur in the generated results, making it difficult to meet the engineering accuracy requirements in the automotive manufacturing scenario.

[0003] In response to the above problems, no effective solution has been proposed yet. Summary of the Invention

[0004] Embodiments of the present invention provide a method and device for generating a three-dimensional parametric model based on artificial intelligence, so as to at least solve the technical problem of low engineering accuracy existing in the application of existing three-dimensional modeling technologies to mechanical components.

[0005] According to one aspect of an embodiment of the present invention, there is provided a method for generating a three-dimensional parametric model based on artificial intelligence. The method includes: receiving natural language description information input by a user for describing the design requirements of a mechanical component, and performing semantic parsing on the natural language description information to obtain structured requirement data; based on the structured requirement data, using a deep generation model to generate an initial form of the mechanical component, and performing geometric construction on the initial form to obtain a basic geometric body of the mechanical component; performing structural optimization and manufacturing feasibility verification on the basic geometric body to generate the three-dimensional parametric model that meets the design requirements; wherein, performing structural optimization and manufacturing feasibility verification on the basic geometric body includes: based on finite element analysis, performing a stress analysis on the basic geometric body, and based on the result of the stress analysis, performing structural optimization on the basic geometric body to obtain the basic geometric body after structural optimization; performing geometric repair and manufacturing feasibility verification on the basic geometric body after structural optimization, wherein the geometric repair includes at least one of the following: surface smoothing processing and topological structure repair, and the manufacturing feasibility verification includes at least one of the following: minimum thickness inspection and tolerance inspection.

[0006] According to another aspect of an embodiment of the present invention, there is also provided a device for generating a three-dimensional parametric model based on artificial intelligence, including: an analysis module configured to receive natural language description information input by a user for describing the design requirements of a mechanical component, and perform semantic parsing on the natural language description information to obtain structured requirement data; a generation module configured to, based on the structured requirement data, use a deep generation model to generate an initial form of the mechanical component; based on the structured requirement data, perform geometric construction on the initial form to obtain a basic geometric body of the mechanical component; perform structural optimization and manufacturing feasibility verification on the basic geometric body to generate the three-dimensional parametric model that meets the design requirements; wherein, the generation module is further configured to: based on finite element analysis, perform a stress analysis on the basic geometric body, and based on the result of the stress analysis, perform structural optimization on the basic geometric body to obtain the basic geometric body after structural optimization; perform geometric repair and manufacturing feasibility verification on the basic geometric body after structural optimization, wherein the geometric repair includes at least one of the following: surface smoothing processing and topological structure repair, and the manufacturing feasibility verification includes at least one of the following: minimum thickness inspection and tolerance inspection.

[0007] In an embodiment of the present invention, natural language description information input by a user for describing the design requirements of a mechanical component is received, and semantic parsing is performed on the natural language description information to obtain structured requirement data; based on the structured requirement data, a deep generation model is used to generate an initial form of the mechanical component, and geometric construction is performed on the initial form to obtain a basic geometric body of the mechanical component; structural optimization and manufacturing feasibility verification are performed on the basic geometric body to generate the three-dimensional parametric model that meets the design requirements; wherein, performing structural optimization and manufacturing feasibility verification on the basic geometric body includes: based on finite element analysis, performing a force analysis on the basic geometric body, and performing structural optimization on the basic geometric body based on the result of the force analysis to obtain the basic geometric body after structural optimization; performing geometric repair and manufacturing feasibility verification on the basic geometric body after structural optimization, wherein the geometric repair includes at least one of the following: surface smoothing processing and topological structure repair, and the manufacturing feasibility verification includes at least one of the following: minimum thickness inspection and tolerance inspection. Through the above solution, the technical problem of low engineering accuracy existing in the existing three-dimensional modeling technology when applied to mechanical components is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0009] Figure 1 is a flowchart of an optional method for generating a three-dimensional parametric model based on artificial intelligence according to an embodiment of the present invention;

[0010] Figure 2 is a flowchart of an optional method for generating a three-dimensional parametric model for automotive tooling design according to an embodiment of the present invention;

[0011] Figure 3 is a flowchart of an optional method for parsing and processing natural language description information according to an embodiment of the present invention;

[0012] Figure 4 is an interface diagram of an optional intelligent dialogue input and reasoning process according to an embodiment of the present invention;

[0013] Figure 5 is a flowchart of an optional method for generating a three-dimensional parametric model based on structured requirement data according to an embodiment of the present invention;

[0014] Figure 6 is an interface diagram of an optional visual preview of a three-dimensional model file according to an embodiment of the present invention;

[0015] Figure 7 It is a flowchart of another alternative method for generating a 3D parametric model based on artificial intelligence according to an embodiment of the present invention;

[0016] Figure 8 It is a flowchart of an alternative method for semantic parsing of natural language description information according to an embodiment of the present invention;

[0017] Figure 9 It is a structural diagram of an alternative device for generating a 3D parametric model based on artificial intelligence according to an embodiment of the present invention;

[0018] Figure 10 It shows a schematic structural diagram of an electronic device suitable for implementing the embodiments of the present disclosure. Detailed implementation manners

[0019] In order to enable those skilled in the art of the present technology to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0021] According to an embodiment of the present invention, a method embodiment of a method for generating a 3D parametric model based on artificial intelligence is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0022] Figure 1 It is a method for generating a 3D parametric model based on artificial intelligence according to an embodiment of the present invention, as Figure 1As shown, the method includes the following steps:

[0023] Step S102: Receive the natural language description information input by the user for describing the design requirements of the mechanical component, and perform semantic parsing on the natural language description information to obtain structured requirement data.

[0024] First, use the pre-trained language model fine-tuned on the domain to perform semantic parsing on the natural language description information to generate a semantic embedding representation; then, based on the semantic embedding representation, use the named entity recognition and dependency syntactic analysis method to extract the parameters of the mechanical component from the natural language description information to form a preliminary parameter set; then, use the retrieval-enhanced generation mechanism to query relevant entry contents from the knowledge graph, and supplement the missing or incomplete parameters in the preliminary parameter set based on the queried relevant entry contents to obtain the complete parameter set; finally, use rule-based logical reasoning to perform compliance verification on the parameters in the complete parameter set to obtain the verified parameter data, where the compliance verification includes at least one of the following: size range, proportional relationship, material suitability, and assembly constraint; finally, convert the verified parameter data into the structured requirement data in the standard format.

[0025] In this embodiment, by introducing a multi-stage natural language understanding and knowledge completion mechanism, not only the accuracy and integrity of extracting design parameters from natural language are improved, but also the consistency and standardization of data are enhanced through rule reasoning, providing high-quality structured input for subsequent 3D modeling.

[0026] Step S104: Based on the structured requirement data, use a deep generation model to generate the initial form of the mechanical component, and perform geometric construction on the initial form to obtain the basic geometric body of the mechanical component.

[0027] First, based on the structured requirement data, use a deep generation model to generate the initial form of the mechanical component. For example, use a variational autoencoder to perform feature encoding on the structured requirement data to generate a latent structure representation that conforms to the prior design constraints; through an adversarial network, perform adversarial training on the latent structure representation that conforms to the prior design constraints to generate a latent structure representation with geometric details; use a diffusion model to perform denoising optimization on the latent structure representation with geometric details to generate the initial form of the mechanical component.

[0028] Next, perform geometric construction on the initial form to obtain the basic geometric body of the mechanical component. For example, use boundary representation to perform geometric construction on the initial form to obtain an initial geometric body; use non-uniform rational B-splines to perform surface modeling on the geometric region corresponding to the free surface of the initial geometric body to obtain the basic geometric body.

[0029] Through the above method, this embodiment realizes the three-dimensional form construction process from rough to fine, which not only improves the rationality of the initial model in terms of structural logic, but also significantly enhances the geometric details and realistic performance.

[0030] Step S106: Perform structural optimization and manufacturing feasibility verification on the basic geometric body to generate the three-dimensional parametric model that meets the design requirements.

[0031] First, based on finite element analysis, perform a force analysis on the basic geometric body, and based on the results of the force analysis, perform structural optimization on the basic geometric body to obtain the basic geometric body with optimized structure. For example, based on the material properties, boundary conditions, and load conditions of the basic geometric body, construct a finite element analysis model; perform mesh division on the finite element analysis model, perform a force analysis on the basic geometric body based on the divided mesh, and solve the stress, strain, and displacement field distributions of the basic geometric body as the results of the force analysis; based on the results of the force analysis, identify the stress concentration areas of the basic geometric body; based on the stress concentration areas, use a topology optimization algorithm to adjust the structural dimensions and reconstruct the material distribution of the basic geometric body to obtain the basic geometric body with optimized structure.

[0032] Then, perform geometric repair and manufacturing feasibility verification on the basic geometric body with optimized structure. For example, use a surface reconstruction algorithm to fit the irregular surface of the basic geometric body with optimized structure, and based on the surface curvature distribution of the fitted basic geometric body, perform surface smoothing on the fitted basic geometric body; use a topology structure repair algorithm to repair the boundary openings, mesh breakages, and self-intersection areas of the basic geometric body after surface smoothing; based on the preset manufacturing process constraint conditions, use a rule matching algorithm to perform the minimum thickness check and the tolerance check on the basic geometric body after topology structure repair to perform manufacturing feasibility verification.

[0033] Through the above method, this embodiment can accurately express the hybrid form with regular structures and free-form surfaces, improving the expression ability and flexibility of modeling.

[0034] The embodiment of the present application also provides a method for generating a three-dimensional parametric model for automotive tooling design. This method realizes the one-stop generation from language requirements to a manufacturable tooling model by integrating natural language interaction, knowledge graph enhancement, deep learning generation, and parametric optimization technologies. Specifically, in this embodiment, based on a natural language-driven design interface, the requirement description input by the user is converted into three-dimensional modeling instructions through semantic parsing; combined with the knowledge graph enhancement technology, a fixture design knowledge base containing positioning accuracy standards and material mechanics parameters is used to automatically complete missing parameters such as clamping force and safety factor to ensure design reliability; and, a hybrid modeling architecture is adopted. Complex structures such as surface frameworks or special-shaped support frameworks are generated through deep learning algorithms, and at the same time, parametric modeling technology is used to ensure the dimensional accuracy of key interfaces such as pin hole positions and cylinder mounting surfaces; finally, a manufacturability optimization closed-loop is constructed, and the model is automatically optimized through assembly interference detection and finite element strength analysis to ensure compliance with body manufacturing process requirements such as welding accessibility and repeat positioning accuracy.

[0035] Figure 2 The flowchart of a method for generating a three-dimensional parametric model for automotive tooling design provided by the embodiment of the present application is shown. As Figure 2 shown, the method includes the following steps:

[0036] Step S202, parse and process the natural language description information.

[0037] Specifically, the process of parsing and processing the natural language description information is as Figure 3 shown, and includes the following steps:

[0038] Step S2022, input the design requirements of the user into the LLM model.

[0039] First, construct and train the LLM model. In this embodiment, a large language model (LLM) is used as the core inference engine, and the training data includes annotated design documents of vehicle body parts, industry standards, and enterprise internal specifications, etc. The model training adopts the method of freezing the underlying parameters and fine-tuning the top-layer parameters, or the method of full-parameter fine-tuning of the model, and combines domain data augmentation technology to improve the model effect. In addition, a distributed search engine such as Elasticsearch or a vector similarity search library Faiss is used to build a vector index, storing part specifications (such as "the guide cone cone angle ≥ 60°"), material properties (such as "the yield strength of aluminum alloy ≥ 200 MPa"), etc., to form a knowledge base. During the inference process, by converting the text input by the user into an embedding vector, relevant entries in the knowledge base are retrieved and used as context input to the LLM, thereby enhancing the model's understanding and reasoning ability of domain knowledge. In addition, dynamic dialogue management is implemented through the Langchain framework to achieve multi-turn interaction and state tracking, such as recording the parameter information confirmed by the user, etc.

[0040] Next, input the design requirements entered by the user into the trained LLM model. The user input can be natural language text or speech. The text input can be, for example, "Generate a guide cone with a height of 50 mm, a bottom diameter of 40 mm, and with chamfers", or it can be a more complex description. The intelligent dialogue input and inference process is as Figure 4 shown. The speech input can be transcribed into text using an ASR tool to achieve multi-language input such as Chinese and English, and the speech source information can be marked, such as the language type and accent type. The LLM model can also automatically obtain historical interaction data for context modeling.

[0041] The data sources include a pre-trained corpus, a domain knowledge base, and user feedback data. The pre-trained corpus can include content such as automotive manufacturing industry forums, technical blogs, patent documents, etc., and data cleaning, noise filtering, and key technical term annotation are required before use. The domain knowledge base can be constructed as a knowledge graph by storing part attribute relationships through Neo4j or Redis Graph, and can be updated in real time to reflect the latest industry standards. In some embodiments, the user feedback data is stored in JSON format, including a timestamp, a user ID, and a context ID, and only the last 5 rounds of conversations are retained to control the length of the context window.

[0042] Step S2024, parse the user input information.

[0043] First, perform text parsing. Extract the key semantic information from the design requirements input by the user through the LLM model to obtain the semantic embedding representation, and use the fine-tuned domain data augmentation model to accurately parse the descriptions related to vehicle body parts. Use named entity recognition (NER) technology to annotate the part names, units, constraint conditions, etc., and establish the relationships between parameters through the dependency syntactic analysis method. For example, in "diameter 50mm", "50mm" is the modifier of "diameter".

[0044] Next, complete the information. Through the RAG (retrieval-augmented generation) mechanism, query the relevant content in the knowledge graph to automatically complete the parameter information not explicitly stated by the user, such as the default thickness, standard size, etc.

[0045] Then, perform logical reasoning. Based on the preset rules, perform constraint solving, including size ranges, proportional relationships, and assembly constraints between parts. If incomplete information or unreasonable parameters are found, prompt the user to supplement the information. In this embodiment, through the intelligent and accurate parameter parsing and completion capabilities, combined with pre-trained language models (such as GPT / BERT) and knowledge base enhancement, the parsing accuracy and integrity are ensured, and missing parameters can be automatically filled.

[0046] Step S2026, data structuring.

[0047] Convert the parsed semantic information into structured requirement data in a standard format, such as JSON or XML format, for use by the subsequent 3D parametric model generation module. An example of the structured requirement data is as follows:

[0048] {

[0049] "part_type": "guide cone",

[0050] "cylinder_radius": 10.0,

[0051] "cylinder_height": 51.0,

[0052] "cone_radius": 34.0,

[0053] "cone_height": 39.0,

[0054] "fillet_radius": 20.0

[0055] }

[0056] In this embodiment, by directly describing the requirements in natural language, automatically parsing and generating structured requirement data, without manual modeling, significantly improves the design convenience.

[0057] Step S204: Generate a 3D parametric model based on the structured requirement data.

[0058] As Figure 5 shown, the method for generating a 3D parametric model based on the structured requirement data includes the following steps:

[0059] Step S2042: Construct a 3D parametric model.

[0060] Constructing a 3D parametric model specifically includes three stages:

[0061] First, construct a deep learning generation module for generating the initial form of complex structures using deep generative models (such as VAE, GAN, diffusion models). The deep learning generation module includes a VAE network, a GAN network, and a diffusion model. The VAE network is used to learn the distribution characteristics of components to generate an initial form that conforms to the prior design. The GAN network is used to generate high-fidelity geometric details, and the diffusion model is used to improve the realism and detail performance of the model through denoising optimization.

[0062] Next, construct a parametric modeling module. The parametric modeling module performs geometric construction based on the structured requirement data using a parametric modeling engine. For example, use boundary representation (B-Rep) to construct basic geometric bodies such as cubes, cylinders, and cones, and use non-uniform rational B-splines (NURBS) for modeling complex surfaces.

[0063] Finally, construct an optimization and verification module. This module combines finite element analysis (FEA) for real-time simulation, calculates the force distribution, optimizes the structural strength, and simultaneously performs geometric repair and manufacturing feasibility verification, including surface smoothing, topology repair, minimum thickness and tolerance inspection, etc.

[0064] This embodiment introduces a multi-level optimization mechanism to ensure that the generated 3D parametric model has good manufacturing feasibility; checks the force condition through finite element analysis, improves the surface quality through geometric optimization, and guarantees manufacturing compliance through dimension verification; provides a real-time interaction and feedback mechanism; if the user input is incomplete or contains logical errors, it can automatically prompt and assist in correction, and combines the visualization preview function to improve the user experience and interaction efficiency.

[0065] Step S2044: Input the data required for modeling.

[0066] The input includes structured requirement data, additional modeling constraints, and a historical 3D model database. The structured requirement data is derived from the natural language processing results of the above steps and includes component types, dimensions, and constraint conditions, etc. The additional modeling constraints are specific requirements input by the user, such as accuracy, tolerance, assembly conditions, etc. The historical 3D model database contains CAD models of standard components (such as STEP, STL formats) that can be used for feature extraction and style transfer.

[0067] The data sources can include the 3D model database of vehicle body parts, manufacturing process data, and existing CAD data. The 3D model database contains standard models, point clouds, and mesh data, and is equipped with parameter attribute annotations. The manufacturing process data stores parameter information related to modeling and can be updated regularly. The existing CAD data can extract key geometric features through reverse engineering to form a feature library that can be called by the modeling module.

[0068] Step S2046: Perform 3D modeling.

[0069] The 3D modeling includes three parts: In the deep learning part, VAEs, GANs, and diffusion models are used to generate the initial form. The VAE is used to learn the design distribution, the GAN is used for high-quality geometry generation, and the diffusion model is used to further optimize the details to improve the authenticity of the model. In the geometric construction part, the parametric modeling engine generates the corresponding geometric model according to the structured requirement data. The basic geometric shape is constructed based on boundary representation (B-Rep), and non-uniform rational B-splines are used to handle complex surface modeling. In the optimization and verification part, physical simulation is combined with FEA to optimize the structural strength; surface smoothing and structure repair are achieved through geometric optimization; dimensional verification is performed to ensure compliance with manufacturing standards, such as aperture and thickness limits.

[0070] This embodiment integrates deep learning and parametric modeling technologies, quickly generates complex structures through VAEs, GANs, and diffusion models, and then accurately restores geometric bodies with parametric modeling, enabling efficient generation of 3D parametric models.

[0071] Step S2048: Output the modeling result.

[0072] The modeling result is output as a standardized 3D model file, such as STL, STEP, IGES formats, and a visualization preview function based on Three.js / WebGL is provided, as Figure 6 shown. Users can choose to download the model or make further modifications.

[0073] This embodiment automatically generates 3D parametric models driven by natural language, greatly improving the efficiency and intelligent level of vehicle body part design.

[0074] Another method for generating a 3D parametric model based on artificial intelligence is also provided in the embodiment of this application, as Figure 7 shown, and this method includes the following steps:

[0075] Step S702: Perform semantic parsing.

[0076] Specifically, the method of semantic parsing is as Figure 8 shown and includes the following steps:

[0077] Step S7022: Receive and fuse multi-modal input information.

[0078] In the embodiment of the present application, not only the natural language text description input by the user is received, but also the auxiliary design materials such as hand-drawn sketches, structural schematic diagrams or historical design drawings are supported to be received synchronously. Through the multi-modal alignment algorithm, the semantic features in the text information and the image information are uniformly embedded and encoded by using the multi-modal pre-trained model, and elements such as part names, dimension parameters, function descriptions, assembly directions and boundary conditions are extracted, providing a multi-source basis for subsequent semantic parsing and modeling.

[0079] Step S7024: Construct a semantic constraint graph.

[0080] Based on the multi-modal design elements (i.e., semantic embedding representation) extracted in Step S7022, the entity recognition and dependency syntactic analysis methods are adopted to extract the subordinate, modifying and constraint relationships between the parameters, and a semantic constraint graph is constructed. The graph represents the design parameters, functional objectives and structural parts with nodes, and represents the logical association and assembly dependency relationships between the parameters with edges, which is used to support the downstream knowledge completion and rule reasoning processes.

[0081] Step S7026: Perform multi-source parameter completion.

[0082] For the parameter nodes with missing or ambiguous information in the semantic constraint graph, based on the multi-source joint completion mechanism, the candidate completion results are extracted in parallel from the standard parts database, knowledge graph, design case database, user historical interaction data and the content generated by the language model. A credibility evaluation model is introduced to score each candidate parameter, and the scoring basis includes indicators such as its consistency with the semantic graph structure, the authority of the source, and the relevance to the task objective. The optimal completion scheme is selected according to the game optimization strategy, and the completion result is filled back into the semantic graph.

[0083] Specifically, after the structured semantic information is extracted, the missing items in the preliminary parameter set are identified, and a multi-source parameter completion process driven by strategies is started. First, a semantic dependency graph with the currently parsed parameters as nodes is constructed, and the dependency relationships, semantic constraints and entity types between the parameters are encoded as a graph structure, so that the subsequent processing process can maintain context consistency. Then, based on this semantic graph structure, multiple knowledge source modules are called simultaneously, including the standard parts database, the semantic enhanced knowledge graph, the user customization preference model and the historical model instance set. During the calling process, according to the confidence and context dependency levels of each node in the semantic dependency graph, the calling priorities of each knowledge source are dynamically determined, so as to achieve adaptive knowledge scheduling.

[0084] To improve the completion quality, this embodiment introduces a graph neural network (GNN) model to perform context completion reasoning on the missing nodes in the semantic graph, learns the design dependencies in historical samples based on multi-hop paths, and makes a preliminary prediction of the completion parameters. On this basis, using the RAG mechanism of the language model, potential completion content is retrieved from the external semantic index, and multiple candidate parameter sets are obtained by fusing the results of the generative and retrieval methods. Subsequently, a hybrid confidence scoring mechanism is used to rank the candidate sets. This score comprehensively considers semantic consistency, knowledge source weights, historical matching rates, structural rationality, and whether material and process constraints are met, and assigns an interpretable source label and adaptation mark to each candidate parameter.

[0085] In addition, a parameter completion decision graph is constructed to transform multiple candidate completion paths into a directed acyclic graph structure. An optimal path search algorithm (such as heuristic A* or Monte Carlo tree search) is used to select the optimal solution path that meets the constraints, and the final completion parameters are injected into the initial parameter set. To enhance the user experience and improve robustness, this application also provides a transparent interactive feedback interface for completion items, supporting users to optionally confirm, replace, or reject the suggested items. At the same time, this feedback information will be used in reverse to update the knowledge source weights and preference models.

[0086] The embodiment of this application, by introducing semantic reasoning based on graph neural networks, a dynamic knowledge scheduling mechanism, candidate parameter confidence ranking, and an optimal path completion algorithm, is different from the existing method of static completion based only on rules or a single model, significantly improving the intelligence, self-adaptability, and accuracy level of parameter completion, and is particularly suitable for complex and multi-constrained mechanical structure design scenarios.

[0087] Step S7028, perform rule-driven logical reasoning.

[0088] Based on the completed semantic constraint graph, load the predefined assembly rule library and design specifications, execute the logical reasoning process, and verify the rationality of the design parameters. The reasoning content includes but is not limited to: dimension range check, component ratio relationship verification, structural interference detection, material adaptability, and assembly constraints. For parameters that violate the rules or paths where the reasoning fails, they will be automatically marked and a prompt will be generated to guide the user to correct or supplement the information.

[0089] Step S7029, output structured requirement data.

[0090] Extract the verified design parameters from the semantic constraint graph, and convert them into a structured data format according to the preset standard model construction rules, such as formats that can be parsed by modeling engines like JSON and XML. This structured output not only retains the complete parameter content but also includes source annotations, credibility scores, and rule verification labels to support the precise construction and parameter traceability of subsequent modeling modules.

[0091] In this embodiment, by introducing multi-modal information fusion, multi-source trustworthy completion, game optimization selection, and graph-driven reasoning mechanisms, the intelligent, accurate, and interpretable analysis of the user's design intent is significantly improved, and problems in the prior art such as relying on single-text input, lacking semantic association modeling, and lacking credibility assessment for parameter completion are solved.

[0092] Step S704, generate a basic geometric body.

[0093] 1) Generate an initial form.

[0094] For example, use a variational autoencoder (VAE) to perform feature encoding on structured requirement data, including dimension parameters, function descriptions, assembly constraints, boundary conditions, and material properties, etc., map it to the latent space, and generate a latent structure representation that meets the design specifications. This latent space is optimized by maximizing the variational lower bound (ELBO) during the training phase to ensure that it captures the latent distribution characteristics of key design elements.

[0095] Subsequently, for the above latent structure representation, use a generative adversarial network (GAN) to enhance geometric details. Specifically, take the output of the VAE as the initial input of the generator. By introducing a multi-scale convolutional module and a residual connection mechanism, the generator can capture local geometric changes and global structural consistency. The discriminator discriminates the generated results based on prior training samples to distinguish real structure samples from generated structure samples. Through repeated iterative adversarial training, the latent structure representation output by the generator gradually has the geometric complexity and detail accuracy required for real mechanical components.

[0096] After completing the adversarial training, to further improve the geometric quality and manufacturability of the initial form, use a diffusion model to denoise and optimize the above latent structure representation. Specifically, introduce a diffusion process based on forward Gaussian noise perturbation to gradually perturb the complex geometric structure into a standard distribution form, and then gradually restore the denoised form through the reverse process. Control variables are introduced in this process, such as the type of target component, the level of geometric complexity, and the manufacturing tolerance requirements, etc., to guide the sampling direction in the diffusion reverse process and enhance the adaptability of the generated form to prior constraints. In addition, this embodiment also adopts a multi-stage fusion strategy to dynamically inject structured semantic labels (such as key dimensions, assembly interface positions, etc.) in the intermediate steps of the diffusion model to improve the fidelity of the final generated structure to the original design intent.

[0097] The initial form output in this embodiment not only has high geometric fidelity but also integrates semantic information and physical constraints, providing a structural basis with complete prior knowledge for subsequent geometric construction and 3D modeling. In addition, through the collaborative optimization of the deep generation model, this embodiment effectively solves the problems of lack of geometric details, poor controllability, and difficulty in adapting to complex prior constraints in traditional structure generation methods, improving the intelligence and practicality of 3D design generation.

[0098] 2) Generate basic geometric bodies.

[0099] First, use Boundary Representation (B-Rep) to perform geometric construction on the initial form. Specifically, based on the geometric feature points, boundary contours, structural skeletons, etc. extracted from the initial form, construct a topological structure description model composed of vertices, edges, and faces. During the construction process, introduce Boolean operations and set operation mechanisms (such as union, difference, intersection) to achieve precise splicing of multiple sub-structures, and adopt boundary legality detection algorithms to ensure topological consistency and closure during the construction process. In addition, to improve the robustness of the geometric model, this embodiment also introduces a surface normal vector consistency correction and boundary closure repair algorithm to avoid modeling failures caused by local discontinuities or floating points in the initial form.

[0100] After completing the geometric construction of the boundary structure, for the free-form surface regions in the initial geometric body, further perform fine surface modeling operations. Specifically, use Non-Uniform Rational B-Splines (NURBS) to perform parametric modeling on the free-form surface regions. This method realizes high-precision modeling of complex curvature changes through the joint definition of control points, weight coefficients, and knot vectors. To improve the adjustability and computational efficiency of the model, adopt an adaptive control point distribution mechanism to automatically adjust the control point density according to the curvature gradient changes in different surface regions of the initial form, increasing the control points in high-curvature regions to improve the fitting accuracy and reducing the control points in flat regions to optimize storage and computational resources.

[0101] In addition, to maintain the consistency between the structural design semantics and geometric modeling, load the functional attributes (such as assembly surfaces, connection holes, reinforcing ribs, etc.) provided by the semantic graph and perform geometric mapping during the modeling process, so that the basic geometric bodies not only have a real morphological appearance but also accurately express the design intent and assembly constraints. For complex transition regions, introduce a multi-segment splicing surface transition algorithm (such as a C1 continuity preservation algorithm) to improve the overall smoothness of the model and avoid sharp corners or breaks at geometric joints.

[0102] The finally obtained basic geometric body has an editable parametric structure, supports the direct invocation of subsequent modules such as geometric optimization, physical simulation, and manufacturing inspection, and can be exported or stored in a standard modeling engine (such as STEP, IGES format), ensuring cross-platform compatibility and reusability of design data. In this embodiment, by integrating boundary representation and free-form surface parametric modeling techniques, high-fidelity and high-degree-of-freedom modeling of complex mechanical structure forms is achieved, effectively making up for the deficiencies of traditional modeling methods in terms of automation and flexible control.

[0103] Step S706: Perform structural optimization and manufacturing feasibility verification on the basic geometric body to generate a three-dimensional parametric model.

[0104] 1) Perform structural optimization on the basic geometric body.

[0105] Specifically, based on the material properties (such as elastic modulus, Poisson's ratio, density, etc.), boundary conditions (such as fixed support, sliding constraint, contact interface, etc.), and load conditions (including various load types such as concentrated load, surface load, body force, inertial load, etc.) of the basic geometric body, a three-dimensional finite element analysis model (FEA) is constructed. This model is based on the B-Rep structure, and the geometric body is discretized into computable unit cells (such as tetrahedron or hexahedron mesh cells) through entity meshing.

[0106] During the mesh generation process, a multi-scale adaptive mesh generation algorithm is adopted. According to the local curvature change of the geometric body and the accuracy requirements of the load application area, the mesh density and element shape are dynamically adjusted to ensure higher simulation accuracy in key parts (such as high stress gradient regions). This embodiment also introduces quality evaluation indicators to control the quality of the mesh, including parameters such as element shape quality, Jacobian determinant, and minimum interior angle, automatically identifying and repairing deformed elements to enhance the stability and convergence of the simulation.

[0107] After the mesh generation is completed, the finite element simulation solution process is executed. Based on a linear or nonlinear static solver, combined with the stiffness matrix equation in solving boundary value problems, the stress field, strain field, and displacement field distributions of the basic geometric body under external loads are calculated. To meet the engineering accuracy requirements, dynamic response analysis (such as modal analysis, harmonic response analysis, etc.) can be further superimposed to obtain the complete mechanical performance of the structure under different working conditions.

[0108] Subsequently, based on the stress analysis results, the stress concentration regions of the basic geometric body are identified. Specifically, based on the maximum principal stress trajectory in the tensor field, the Von Mises equivalent stress distribution, and the structural gradient variability, clustering analysis or heat map distribution is used to identify high-risk regions and form a stress risk map.

[0109] After identifying the high-stress regions, topology optimization is carried out. Topology optimization algorithms based on the density method (SIMP) or the level set method are used to reconstruct the structural dimensions and material distribution of the geometry. The optimization objectives can be set flexibly, including minimizing structural compliance, maximizing stiffness, controlling the weight / volume ratio, increasing modal frequencies, etc. The constraint conditions include engineering requirements such as maximum stress limits, manufacturing area constraints (such as non-removable regions), and support structure strength.

[0110] To improve the optimization convergence efficiency and the manufacturability of the results, this embodiment introduces a multi-stage optimization strategy. First, coarse-grained optimization is performed to quickly converge to the general shape, and then fine reconstruction is carried out to restore the edge details. At the same time, a manufacturability filter is constructed in combination with manufacturing process limitations (such as the minimum wall thickness of 3D printing, milling accessibility, material orientation, etc.) to correct the structural units that do not conform to the manufacturing rules in real time.

[0111] The basic geometry after structural optimization output in this embodiment not only realizes the reasonable distribution of the load path and maximizes the material utilization rate in terms of mechanical properties, but also takes into account engineering constraints and manufacturing adaptability, significantly improving the reliability, economy, and feasibility of the structure.

[0112] 2) Geometric repair is performed on the basic geometry after structural optimization.

[0113] Problems such as topological fractures, surface distortions, hole cracks, and self-intersections often occur during the structural optimization process. Therefore, a geometric repair process needs to be executed to restore a usable model. First, a surface reconstruction algorithm is used to fit and reconstruct the discrete surface point set of the optimized geometry. Specifically, based on the point cloud or triangular mesh derived from the topology optimization results, an algorithm based on Poisson surface reconstruction or signed distance field (SDF) interpolation is called to generate a continuous and smooth implicit surface. To improve the reconstruction quality, a curvature-aware sampling mechanism is introduced to densify the sampling points in the high-curvature regions of the boundary and reduce the reconstruction error.

[0114] After the surface fitting is completed, surface smoothing is further performed based on the surface curvature distribution. During this process, an adaptive Laplacian smoothing algorithm based on Gaussian curvature and mean curvature is used to eliminate redundant sharp corners, mesh jaggedness, and concave structures, while maintaining the invariance of the original structural functional features. To avoid geometric distortion caused by over-smoothing, a boundary protection mechanism is used to fix the positions of key feature lines and assembly interface points, and a local iteration strategy is adopted to control the smoothing range and intensity.

[0115] After completing surface reconstruction and smoothing, enter the topological structure repair stage. First, detect the topological consistency of the geometry, including open boundary recognition, face overlap, self-intersection location, etc. Based on topological graph matching and Euler characteristic number consistency analysis, identify topological defects existing in the geometric structure. Subsequently, call the automatic topological repair module, and use Hole Filling, Edge Stitching, and Mesh Merging technologies to close the boundary and repair the structural continuity to ensure that the final model is a closed manifold (watertight model).

[0116] In addition, to ensure that the model meets the manufacturing requirements, manufacturing constraint information is synchronously introduced in the geometric repair process, and a manufacturability pre-inspection based on rule matching is performed. The rule library includes but is not limited to: minimum wall thickness requirements (such as ≥3mm for castings, ≥1mm for metal 3D printing); structural scale limitations such as hole diameter / slot width / rib spacing, etc.; tolerance band definition (variation within ±0.1mm); manufacturing limits for chamfers, fillets, and transition radii.

[0117] Finally, based on the AABB bounding box and spatial index, scan the geometry region by region, and perform fast matching analysis based on the IF-THEN rule engine to generate a warning list and correction suggestions. Highlight the regions that do not meet the rules for subsequent user interaction or automatic fine-tuning.

[0118] 3) Perform a manufacturability check on the repaired basic geometry.

[0119] To ensure that the geometric model can be actually machined or printed, a comprehensive manufacturability check is performed after completing geometric repair. First, load the set of constraint parameters matching the target manufacturing process (such as injection molding, CNC machining, SLM metal printing, etc.), including material parameters, dimensional constraints, machining directionality, support requirements, etc.

[0120] First, perform a geometric scale check. Use voxelization scanning and local slicing analysis methods to determine whether there are geometric features in the model that violate the minimum thickness, too small gaps, extremely small chamfers, etc. Based on the local voxel granularity, calculate the wall thickness distribution map through continuous layer analysis, and accurately determine whether the minimum feature size is within the machinable range in combination with the Euclidean distance transformation.

[0121] Next, perform a tolerance consistency analysis. For key assembly positions, positioning holes, mating surfaces, etc., load the tolerance specification model, perform a fit grade check (such as H7 / g6), and combine the function information provided by the CAD model annotation and semantic graph to ensure that the interface dimensions between parts meet the assembly requirements. For multi-part assemblies, virtual assembly pre-analysis can also be performed to determine whether there is interference or insufficient clearance.

[0122] Finally, optionally, manufacturing process matching verification can also be performed. Specifically, a process matching map is constructed to evaluate the suitability between each feature region and the manufacturing equipment / material properties. For example, based on the projection direction, determine whether there are internal cavity features that cannot be milled, detect the support angle region required for metal additive manufacturing, or analyze whether there is a risk of deformation in thin-walled structures, etc. The scoring results are presented in the form of a heat map, and machining suggestions are automatically generated, such as adjusting the wall thickness, optimizing the draft direction, rearranging stiffeners, etc.

[0123] This application also provides a generating device for a three-dimensional parametric model based on artificial intelligence, as Figure 9 shown, including: a parsing module 92, configured to receive natural language description information input by a user for describing the design requirements of a mechanical component, and perform semantic parsing on the natural language description information to obtain structured requirement data; a generating module 94, configured to generate an initial form of the mechanical component based on the structured requirement data by using a deep generation model; perform geometric construction on the initial form based on the structured requirement data to obtain a basic geometric body of the mechanical component; perform structural optimization and manufacturing feasibility verification on the basic geometric body to generate the three-dimensional parametric model that meets the design requirements; wherein, the generating module is further configured to: perform a force analysis on the basic geometric body based on finite element analysis, and perform structural optimization on the basic geometric body based on the result of the force analysis to obtain the basic geometric body after structural optimization; perform geometric repair and manufacturing feasibility verification on the basic geometric body after structural optimization, wherein the geometric repair includes at least one of the following: surface smoothing treatment and topological structure repair, and the manufacturing feasibility verification includes at least one of the following: minimum thickness inspection and tolerance inspection.

[0124] It should be noted that: the generating device for a three-dimensional parametric model based on artificial intelligence provided in the above embodiments is only illustrated by dividing the above functional modules. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the generating device for a three-dimensional parametric model based on artificial intelligence provided in the above embodiments and the embodiments of the generating method for a three-dimensional parametric model based on artificial intelligence belong to the same concept, and the specific implementation process is detailed in the method embodiments and will not be repeated here.

[0125] Figure 10 shows a schematic structural diagram of an electronic device suitable for implementing the embodiments of the present disclosure. It should be noted that Figure 10 the shown electronic device is only an example and should not bring any limitation to the functions and usage scope of the embodiments of the present disclosure. As Figure 10As shown, the electronic device includes a processor 1001, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 1002 or the program loaded from the storage section 1008 into the random access memory (RAM) 1003. In the RAM 1003, various programs and data required for system operation are also stored. The CPU 1001, ROM 1002, and RAM 1003 are connected to each other via a bus 1004. The input / output (I / O) interface 1005 is also connected to the bus 1004.

[0126] The processor 1001 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 1001 executes the various methods and processes described above, such as the model training method.

[0127] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, etc.; an output section 1007 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed so that a computer program read from it can be installed into the storage section 1008 as needed.

[0128] The above description is only the preferred embodiment of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A method for generating a three-dimensional parametric model based on artificial intelligence, characterized in that, Including: Receiving natural language description information input by a user for describing the design requirements of a mechanical component, and performing semantic parsing on the natural language description information to obtain structured requirement data; Based on the structured requirement data, using a deep generation model to generate an initial form of the mechanical component, and performing geometric construction on the initial form to obtain a basic geometric body of the mechanical component; Performing structural optimization and manufacturing feasibility verification on the basic geometric body to generate the three-dimensional parametric model that meets the design requirements; Among them, performing structural optimization and manufacturing feasibility verification on the basic geometric body includes: based on finite element analysis, performing a force analysis on the basic geometric body, and based on the results of the force analysis, performing structural optimization on the basic geometric body to obtain the basic geometric body after structural optimization; performing geometric repair and manufacturing feasibility verification on the basic geometric body after structural optimization, where the geometric repair includes at least one of the following: surface smoothing treatment and topological structure repair, and the manufacturing feasibility verification includes at least one of the following: minimum thickness inspection and tolerance inspection.

2. The method according to claim 1, wherein Based on the structured requirement data, using a deep generation model to generate an initial form of the mechanical component, including: Using a variational autoencoder to perform feature encoding on the structured requirement data to generate a latent structure representation that conforms to prior design constraints; Through an adversarial network, performing adversarial training on the latent structure representation that conforms to prior design constraints to generate a latent structure representation with geometric details; Using a diffusion model to perform denoising optimization on the latent structure representation with geometric details to generate the initial form of the mechanical component.

3. The method according to claim 1, wherein Performing geometric construction on the initial form to obtain a basic geometric body of the mechanical component, including: Using boundary representation to perform geometric construction on the initial form to obtain an initial geometric body; Using non-uniform rational B-splines to perform surface modeling on the geometric regions corresponding to the free surfaces of the initial geometric body to obtain the basic geometric body.

4. The method according to claim 1, wherein Performing semantic parsing on the natural language description information to obtain structured requirement data, including: Using a pre-trained language model fine-tuned in the domain to perform semantic parsing on the natural language description information to generate a semantic embedding representation; Based on the semantic embedding representation, using named entity recognition and dependency syntactic analysis methods to extract the parameters of the mechanical component from the natural language description information to form a preliminary parameter set; Using a retrieval-enhanced generation mechanism to query the relevant entry content corresponding to the mechanical component from a preset knowledge graph, and based on the queried relevant entry content, supplementing the missing or incomplete parameters in the preliminary parameter set to obtain a complete parameter set; Using rule-based logical reasoning to perform compliance verification on the parameters in the complete parameter set to obtain a verified parameter set, where the compliance verification includes verifying at least one of the following: size range, proportional relationship, material compatibility, and assembly constraints; Performing standard format conversion on the verified parameter set to obtain the structured requirement data.

5. The method according to claim 1, wherein Based on finite element analysis, perform a force analysis on the basic geometric body, and based on the results of the force analysis, optimize the structure of the basic geometric body to obtain the basic geometric body with an optimized structure, including: Construct a finite element analysis model based on the material properties, boundary conditions, and load conditions of the basic geometric body; Perform mesh division on the finite element analysis model, perform a force analysis on the basic geometric body based on the divided mesh, and solve the stress, strain, and displacement field distributions of the basic geometric body as the results of the force analysis; Based on the results of the force analysis, identify the stress concentration areas of the basic geometric body; Based on the stress concentration areas, use a topology optimization algorithm to adjust the structural dimensions and reconstruct the material distribution of the basic geometric body to obtain the basic geometric body with an optimized structure.

6. The method according to claim 5, wherein Perform geometric repair and manufacturing feasibility verification on the basic geometric body with an optimized structure, including: Use a surface reconstruction algorithm to fit the irregular surface of the basic geometric body with an optimized structure, and based on the surface curvature distribution of the fitted basic geometric body, perform surface smoothing on the fitted basic geometric body; Use a topological structure repair algorithm to repair the boundary openings, mesh breakages, and self-intersection areas of the basic geometric body after surface smoothing; Based on preset manufacturing process constraint conditions, use a rule matching algorithm to perform the minimum thickness check and the tolerance check on the basic geometric body after topological structure repair to perform manufacturing feasibility verification.

7. An apparatus for generating a three-dimensional parametric model based on artificial intelligence, characterized in that, Including: A parsing module configured to receive natural language description information input by a user for describing the design requirements of a mechanical component, and perform semantic parsing on the natural language description information to obtain structured requirement data; A generation module configured to, based on the structured requirement data, use a deep generation model to generate an initial form of the mechanical component, perform geometric construction on the initial form to obtain the basic geometric body of the mechanical component; perform structural optimization and manufacturing feasibility verification on the basic geometric body to generate the three-dimensional parametric model that meets the design requirements; Wherein, the generation module is further configured to: based on finite element analysis, perform a force analysis on the basic geometric body, and based on the results of the force analysis, optimize the structure of the basic geometric body to obtain the basic geometric body with an optimized structure; perform geometric repair and manufacturing feasibility verification on the basic geometric body with an optimized structure, wherein the geometric repair includes at least one of the following: surface smoothing and topological structure repair, and the manufacturing feasibility verification includes at least one of the following: minimum thickness check and tolerance check.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein when the program runs, it controls the device where the computer-readable storage medium is located to execute the method according to any one of claims 1 to 6.

9. A computer device, characterized in that, Including: A memory and a processor, The memory stores a computer program; The processor is configured to execute the computer program stored in the memory, and when the computer program runs, the processor is caused to execute the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images