Hierarchical text-driven three-dimensional modeling method, device and medium

Abstract

The invention discloses a hierarchical text-driven three-dimensional modeling method and device and a medium, belongs to the technical field of engineering software, and is used for solving the technical problems that an existing CAD modeling method depends on manual interface execution, the design threshold is high, the modeling process is tedious and low in efficiency, and the development cycle of a product is easily affected. The method comprises the steps that received product design language demand information is subjected to analysis processing of related functional components and component topological relations through a demand feature decomposition engine, and structured product component information is obtained; according to the original demand information of the user and the product component information, deducing specific geometric parameters of each component of the product to obtain product structure parameter information; performing conversion processing of corresponding execution parameters on the execution command of the CAD platform to obtain a product parameterized modeling script; and performing three-dimensional model construction execution processing on the product parameterized modeling script to generate a product three-dimensional model.

Description

Technical Field

[0001] This application relates to the field of engineering software, and more particularly to a hierarchical text-driven 3D modeling method, device, and medium. Background Technology

[0002] In the product design and development process of the manufacturing industry, CAD (Computer-Aided Design) technology is widely used in the construction and simulation analysis of 3D models. Traditional CAD modeling relies on designers manually performing a series of modeling operations (such as extrusion, rotation, chamfering, arraying, etc.) through a graphical interface. The entire process requires operators to have rich modeling experience and proficient software operation skills. This not only raises the design threshold and limits the participation of non-professional users, but also leads to a cumbersome and inefficient modeling process. Especially when facing complex structures or multiple iterations, it is time-consuming and labor-intensive, seriously affecting the product development cycle.

[0003] Therefore, there is an urgent need for an intelligent modeling system that can efficiently parse natural language requirements and support precise parameter adjustments, in order to solve the problems of high threshold, low efficiency and unstable quality of traditional interactive modeling, while making up for the shortcomings of current text-driven modeling in terms of interpretability and editability. Summary of the Invention

[0004] This application provides a hierarchical text-driven 3D modeling method, device, and medium to solve the following technical problems: existing CAD modeling methods rely heavily on manual interface execution, have high design thresholds, and involve cumbersome and inefficient modeling processes, which can easily affect product development cycles.

[0005] The embodiments of this application adopt the following technical solutions: On one hand, embodiments of this application provide a hierarchical text-driven 3D modeling method, including: using a requirement feature decomposition engine to parse and process received product design language requirement information regarding functional components and component topological relationships to obtain structured product component information; based on the user's original requirement information and the product component information, performing parameter deduction processing on each component of the product for potential geometric parameters and potential related structural components in specific geometry to obtain product structural parameter information; using the user's original requirement information, the structured product component information, and the product structural parameter information, performing conversion processing on the execution commands of the CAD platform for process content and process design parameters to obtain a product parametric modeling script; and performing 3D model construction execution processing on the product parametric modeling script based on key dimensions and key component types to generate a product 3D model.

[0006] This application's embodiments are centered on cloud-based 3D CAD architecture. By constructing a hierarchical parsing process, an editable parameter feedback mechanism, and a multi-level structured collaborative training system, it achieves efficient and controllable generation from fuzzy text descriptions to precise 3D models. Moreover, users do not need to master complex modeling commands or possess professional design experience; they can complete 3D model construction simply through natural language descriptions. This allows junior designers, non-technical personnel, and even ordinary users to participate in the product design process, promoting the widespread application of CAD technology. Furthermore, by combining LLM (Limited Language Management) to convert "text instructions" into "modeling scripts" that can be recognized and executed by the CAD system, a complete technical link of "text instructions - modeling scripts - CAD models" is established, enabling rapid generation of CAD models and significantly improving modeling efficiency. Users can not only make overall adjustments through text feedback but also directly edit component structures and geometric parameters, achieving a closed-loop iteration of "semantic interpretability and parameter editability," balancing ease of use and accuracy.

[0007] In one feasible implementation, before the received product design language requirement information is parsed and processed by the requirement feature decomposition engine to obtain structured product component information by analyzing the relevant functional components and component topology relationships, the method further includes: constructing a text-driven 3D modeling task before the requirement feature decomposition engine performs analysis; wherein the 3D modeling task has several training samples; configuring key fields in the JSON object of each training sample to obtain a fine-tuning subset of the 3D modeling task in different input and output combinations; wherein the key fields include: unique identifier, application domain, difficulty level, user-input natural language modeling requirements, decomposed list of functional components, geometric parameters of each component, and executable parametric modeling script; using the fine-tuning subset, the requirement feature decomposition engine, geometric parameter derivation engine, and parametric modeling script engine are trained based on temporal consistency features to obtain trained component engines.

[0008] In one feasible implementation, a requirement feature decomposition engine is used to parse the received product design language requirement information for relevant functional components and component topology relationships to obtain structured product component information. Specifically, this includes: receiving the product design language requirement information input by the user in natural language form through a user interface; parsing the product design language requirement information for semantic grouping, structural independence, and text content under state constraints using a large language model to obtain preliminary product parsing information; and decomposing the key structural features covering all requirements of the components in the preliminary product parsing information into functional components based on the trained requirement feature decomposition engine. The process involves: identifying key structural features of the product; using a trained requirement feature decomposition engine to perform structural constraint decomposition on the core spatial constraint features related to the topological relationships of the requirement components and other components in the preliminary product analysis information, thereby obtaining the core spatial constraint features of the interconnected components; using the trained requirement feature decomposition engine and based on the core spatial constraint features, performing feature requirement decomposition on the preliminary product analysis information to determine whether additional link structures are needed, thereby obtaining additional link structure features; and marking the key structural features, core spatial constraint features, and additional link structure features as component units of the product's basic structural units to obtain the product component information.

[0009] In one feasible implementation, based on the user's original demand information and the product component information, parameter inference processing is performed on each component of the product for potential geometric parameters and potential related structural components in specific geometry to obtain product structural parameter information. Specifically, this includes: parsing the user's original demand information into semantic text using a large language model to obtain product demand and corresponding relationship text information; using a trained geometric parameter inference engine, and based on the product demand and corresponding relationship text information, performing data recognition processing on the specific geometric parameters of each component in the product component information to obtain identified intuitive structural parameter information; wherein, the specific geometric parameters include: geometric type parameters, size parameters, and three-dimensional spatial coordinates; based on the intuitive structural parameter information, and through the geometric parameter inference engine, performing data inference and simulation prediction on the potential geometric parameters and potential related structural components required by each component in the product component information under relevant modeling requirements to obtain potential structural parameter information; wherein, the potential structural parameter information is structural parameter information that has not been directly identified; and combining the intuitive structural parameter information and the potential structural parameter information based on the same product modeling parameters to generate the product structural parameter information.

[0010] In one feasible implementation, the user's original requirement information, the structured product component information, and the product structural parameter information are used to transform the execution commands of the CAD platform into process content and process design parameters to obtain a product parametric modeling script. Specifically, this includes: using a large language model to identify product component types and construct component relationship graphs from the user's original requirement information, the product component information, and the product structural parameter information, and extracting corresponding parameter association binding information and structural constraint relationship information; using the engineering design rule library of the CAD platform to perform semantic rule matching processing on the parameter association binding information and structural constraint relationship information using a rule engine to obtain specific design instructions and derived design parameters; using a CAD command-parameter mapping table to perform parameter slot mapping processing on the specific design instructions and derived design parameters, and based on a constraint solver, settling the mapped constraint relationship parameter group to ensure that all parameter values ​​satisfy product engineering constraints, resulting in a settled modeling process table; and using a trained parametric modeling script engine to generate command scripts from the process content and process design parameters in the modeling process table to obtain the product parametric modeling script.

[0011] In one feasible implementation, the product parametric modeling script is processed to construct a 3D model based on key dimensions and key component types to generate a 3D product model. Specifically, this includes: splitting the product parametric modeling script into atomic and ordered components to obtain a modeling instruction sequence; generating a feature dependency graph based on the parent-child relationships and geometric dependencies in the modeling instruction sequence; automatically applying virtual assembly constraints and pre-resolving spatial conflicts for each component of the product according to the modeling instruction sequence and the feature dependency graph to obtain an initial 3D model; and verifying the initial 3D model based on key dimensions and key component types using key semantic features corresponding to the user's original requirements, product component information, and product structural parameter information, and obtaining the final 3D product model based on the verification results.

[0012] In one feasible implementation, after the product parametric modeling script is processed to build a 3D model based on key dimensions and key component types to generate a 3D product model, the method further includes: if the user's judgment result of the product 3D model is that the requirements are met, then the product 3D model is directly fed back to the user; if the user's judgment result of the product 3D model is that the requirements are not met, then the newly added or deleted content text is input into the user interaction interface, and the 3D model construction process is re-executed until the user's judgment result of the new product 3D model is that the requirements are met.

[0013] In one feasible implementation, if the user's judgment result of the product 3D model is that the requirements are not met, the model configuration items of the relevant component information and geometric parameters in the product 3D model are modified as needed, and the script regeneration and model reconstruction are automatically triggered to obtain the adjusted product 3D model.

[0014] Secondly, embodiments of this application also provide a hierarchical text-driven 3D modeling device, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to execute a hierarchical text-driven 3D modeling method as described in any of the above embodiments.

[0015] Thirdly, embodiments of this application also provide a non-volatile computer storage medium, which is a non-volatile computer-readable storage medium storing at least one program, each program including instructions, which, when executed by a terminal, cause the terminal to execute a hierarchical text-driven 3D modeling method as described in any of the above embodiments.

[0016] This application provides a hierarchical text-driven 3D modeling method, device, and medium. Compared with the prior art, the embodiments of this application have the following beneficial technical effects: 1. This application takes cloud-based 3D CAD as its core and achieves efficient and controllable generation from fuzzy text descriptions to accurate 3D models by constructing a hierarchical parsing process, an editable parameter feedback mechanism, and a multi-level structured collaborative training system.

[0017] 2. Lowering the modeling threshold: Users do not need to master complex modeling operation commands or have professional design experience. They can complete the construction of 3D models simply by describing them in natural language. This allows junior designers, non-technical personnel, and even ordinary users to participate in the product design process, promoting the widespread application of CAD technology.

[0018] 3. Improve modeling efficiency: This invention combines LLM to convert "text instructions" into "modeling scripts" that can be recognized and executed by the CAD system. Through the complete technical link of "text instructions - modeling scripts - CAD models", it realizes the rapid generation of CAD models and greatly improves modeling efficiency.

[0019] 4. Supports refined human-computer collaborative modeling: Users can not only make overall adjustments through text feedback, but also directly edit component structures and geometric parameters to achieve a closed-loop iteration that is "semantically interpretable and parameter editable," balancing ease of use and accuracy. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating a hierarchical text-driven 3D modeling method provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of a hierarchical text-driven 3D modeling device provided in an embodiment of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0022] This application provides a hierarchical text-driven 3D modeling method that can solve the following technical problems: 1) Lack of structured decomposition in the modeling process: Most methods attempt to directly map text into modeling scripts or 3D meshes end-to-end, lacking intermediate semantic parsing and functional component decomposition processes, resulting in weak understanding of complex requirements and poor controllability of modeling results.

[0023] 2) Difficulty in accurately adjusting the model after generation: When the generated model does not meet the user's expectations, the user can only provide feedback by re-entering or modifying the original text instructions (such as "raise the chair a little higher"). However, such vague expressions are prone to semantic ambiguity, and LLM has difficulty accurately understanding the intention to modify, often leading to unexpected changes or requiring multiple trials and errors, which seriously affects the user experience and modeling efficiency.

[0024] 3) Lack of parameter persistence and editability mechanism: Existing systems usually treat the modeling process as a "one-time black box". The generated model parameters are invisible and uncontrollable. Users cannot directly intervene in the component structure or geometric dimensions, and cannot achieve refined human-computer collaborative modeling.

[0025] This application provides a hierarchical text-driven 3D modeling method, such as... Figure 1 As shown, the hierarchical text-driven 3D modeling method specifically includes steps S101-S104: It should be noted that before the requirement feature decomposition engine performs analysis, a text-driven 3D modeling task is constructed. This task has several training samples. Then, key fields are configured in the JSON object of each training sample to obtain a fine-tuned subset of the 3D modeling task for different input and output combinations. These key fields include: a unique identifier, application domain, difficulty level, user-input natural language modeling requirements, a list of decomposed functional components, the geometric parameters of each component, and an executable parametric modeling script. Finally, using this fine-tuned subset, the requirement feature decomposition engine, geometric parameter derivation engine, and parametric modeling script engine are trained based on temporal consistency features, resulting in the trained component engines.

[0026] As a feasible implementation method, the fine-tuning of the hierarchical text-driven 3D modeling system in this application requires a high-quality, structured, manually annotated training dataset. This dataset is designed for text-driven 3D modeling tasks, covers multiple design fields (such as everyday furniture, mechanical structures, household items, etc.), and is organized in a unified format to ensure the consistency and scalability of the training data for each module.

[0027] Each training sample is a JSON object containing the following key fields: { "sample_id": "Unique identifier", "domain_label": "Application domain (e.g., 3D modeling - everyday furniture)" "difficulty": "Difficulty level (easy / medium / complex)" "modeling_requirement": "Natural language modeling requirements input by the user", "modeling_components": "List of disassembled functional components (including spatial constraints)" "component_parameters": "The geometry, dimensions, and 3D coordinates of each component", "modeling_script": "Executable parametric modeling script (such as CrownScript)".

[0028] That is, in a training data example of everyday furniture: { "sample_id": "3d-mod-005", "domain_label": "3D Modeling - Everyday Furniture", "modeling_requirement": "Generate a minimalist solid wood dining chair model: square solid wood seat cushion, backrest, and 4 solid wood legs". "modeling_components": [ Seat cushion: rectangular structure, no additional decoration. "Straight backrest: Vertical rectangular structure, fixed perpendicularly to the rear end of the seat cushion (90° angle), with the bottom surface touching the rear edge of the top surface of the seat cushion, and its width aligned with the center of the seat cushion." Solid wood chair legs (4 pieces): cylindrical structure, symmetrically distributed along the center of the seat cushion bottom surface, with the top surface completely flush with the bottom surface of the seat cushion (without gaps), supporting the seat cushion. "component_parameters": {...}, "modeling_script": " <script>...< / script> " } In other words, this fine-tuning subset serves as a basic resource, and through different input-output combinations, it is used to construct fine-tuning subsets for each of the three core modules (demand feature decomposition engine, geometric parameter derivation engine, and parametric modeling script engine), thereby achieving phased and specialized training.

[0029] During the fine-tuning of the large language model, this system adopts an architecture design of "large model + phased execution of the thinking chain," decomposing the complex text-to-3D modeling task into multiple logically progressive sub-tasks, which are collaboratively completed by the requirement feature decomposition engine, the geometric parameter inference engine, and the parametric modeling script engine, respectively. These three engine modules are essentially instantiations of the same basic large model at different inference stages. That is, through targeted fine-tuning strategies, they achieve high-precision and highly controllable output results within their respective responsibilities.

[0030] S101. Through the requirement feature decomposition engine, the received product design language requirement information is parsed and processed to obtain structured product component information by analyzing the relevant functional components and component topology relationships. That is, the received natural language requirements are semantically parsed to identify the functional goals and structural intentions, and decomposed into several functional components and their topology relationships (such as "cup body", "cup lid", "sealing ring", "threaded connection", etc.), forming structured component information.

[0031] Specifically, the process begins by receiving product design language requirements from users in natural language via a user interface. That is, the interface receives design requirements input by users in natural language (e.g., "Design a cylindrical water cup with a lid, 15cm high and 7cm in diameter") and passes these requirements to the subsequent processing module. Then, a large language model is used to perform semantic grouping, structural independence, and text content parsing under state constraints to obtain preliminary product analysis information.

[0032] Furthermore, based on the trained requirement feature decomposition engine, the key structural features related to all requirements of the covered components in the preliminary product analysis information are decomposed into functional components to obtain the key structural features of the product.

[0033] Furthermore, based on the trained demand feature decomposition engine, the structural constraint decomposition of the core spatial constraint features of the demand components and other components in the preliminary product analysis information is performed to obtain the product core spatial constraint features of the interrelated components.

[0034] Furthermore, based on the trained requirement feature decomposition engine and the core spatial constraint features of the product, the preliminary product parsing information is further decomposed into feature requirements regarding whether additional link structures are needed, thus obtaining additional link structure features.

[0035] Furthermore, the key structural features, core spatial constraint features, and additional link structural features of the product are labeled as components of the product's basic structural units to obtain product component information.

[0036] In one embodiment, a requirement feature decomposition engine is used to automatically break down the user's natural language description of requirements into a set of semantically clear, structurally independent functional components with assembly constraints, serving as the basic units for subsequent modeling. For example: { "instruction": "Based on user modeling requirements, break down the modeling into 5-15 independently modelable core components, which must meet the following requirements: 1. Components cover all key structures of the requirements; 2. Component descriptions must include core spatial constraints with other components; 3. Components can be directly combined through constraints without additional connection structures!" "input": "<modeling_requirement> ", "output": "<modeling_components> "}

[0037] As a feasible implementation method, the user first has a modeling requirement, such as: "Model a wooden chair for me." The requirement feature decomposition engine will analyze the above requirement, and the large model LLM will decompose the chair into components, such as the chair consisting of 4 chair legs, a backrest, and a seat. (Note that the output process is completed automatically by the LLM). In this way, the LLM will generate the following content: ["Seat cushion: cuboid structure, no additional decoration"; "Straight backrest: vertical cuboid structure, fixed perpendicularly to the rear end of the seat cushion (90° angle), bottom surface attached to the rear edge of the top surface of the seat cushion, width direction aligned with the center of the seat cushion"; "Solid wood chair legs (4): cylindrical structure, symmetrically distributed along the center of the bottom surface of the seat cushion, top surface completely attached to the bottom surface of the seat cushion (no gap), supporting the seat cushion"]. Finally, the components of the product's basic structural units are marked to obtain the product component information.

[0038] S102. Based on the user's original requirements and product component information, perform parameter deduction processing on the potential geometric parameters and potential related structural components of each component of the product to obtain product structural parameter information.

[0039] Specifically, the user's original demand information is first processed through semantic text parsing using a large language model to obtain product demand and corresponding text information.

[0040] Furthermore, through the trained geometric parameter inference engine, and based on product requirements and corresponding textual information, the specific geometric parameters of each component in the product component information are identified and processed to obtain intuitive structural parameter information. These specific geometric parameters include: geometric type parameters, size parameters, and three-dimensional spatial coordinates.

[0041] In one embodiment, based on known user product requirements and functional components, the specific geometric parameters of each component can be deduced, including: geometric type, dimensions (length, width, height, diameter, etc.), and three-dimensional spatial coordinates (x, y, z), forming a complete parameterized definition, for example: {"instruction": "Based on user modeling requirements and component information, estimate the modeling parameters for each component, and specify key information such as the component's geometry type and 3D coordinates!",} "input": { "user_require": "<modeling_requirement> ", "model_components": "<modeling_components> " }, "output": "<component_parameters> "} In other words, the input to the geometric parameter inference engine is the output of the requirement feature decomposition engine, which specifies the components that make up the chair. Then, LLM is used to generate the 3D dimensions of the components, with the JSON text mentioned above as the input. Finally, specific geometric parameter data recognition processing is performed on each component, outputting the intuitive structural parameter information.

[0042] Furthermore, based on the intuitive structural parameter information, and through a geometric parameter deduction engine, the potential geometric parameters required for each component in the product component information, as well as the potential related structural components, are used for data deduction and simulation prediction under the relevant modeling requirements to obtain potential structural parameter information. Among these, potential structural parameter information refers to structural parameters that are not directly identified.

[0043] Furthermore, the intuitive structural parameter information and the potential structural parameter information are combined under the same product modeling parameters to generate product structural parameter information.

[0044] As a feasible implementation method, a geometric parameter derivation engine needs to be developed first. This engine should be able to deduce new geometric parameters based on known geometric constraints and design rules. The engine should be able to handle factors such as geometric transformations, dimensional tolerances, and material properties. It should also be able to generate 3D models or 2D engineering drawings. Next, product component information is analyzed to determine the potential geometric parameters required for each component. Using the geometric parameter derivation engine, these potential geometric parameters are predicted based on the known geometric parameters of the components and design rules. For potentially related structural components, their geometric relationships need to be analyzed, and their potential geometric parameters are predicted. The deduced potential geometric parameters are then stored in a database. Finally, the intuitive structural parameter information and potential structural parameter information are combined. Database queries or data processing software are used to merge parameters from different sources, ensuring that all parameters meet the requirements of product design and manufacturing.

[0045] S103. Using the user's original requirements information, structured product component information, and product structural parameter information, the execution commands of the CAD platform are transformed into relevant process content and process design parameters to obtain the product parametric modeling script.

[0046] Specifically, it is also necessary to use a large language model to process the user's original demand information, product component information, and product structural parameter information to identify product component types and construct a component relationship graph, and extract the product's parameter association binding information and structural constraint relationship information accordingly.

[0047] In one embodiment, a design intent dictionary related to the product domain needs to be pre-defined (e.g., "weight reduction," "stiffness improvement," "volume minimization," "streamlined"). The semantic analysis module extracts keywords from the requirements described in natural language and categorizes them into structural goals, performance goals, and constraints. For example, "as light as possible while ensuring a load capacity of 500N" is parsed as: constraint (load capacity ≥ 500N), structural goal (mass minimization). Then, a component relationship graph is constructed, converting the input component list (e.g., "base, bracket, flange") into a hierarchical relationship diagram. Based on the component name, the system automatically associates its standard interface features (e.g., mating surfaces, bolt hole arrays) and parent-child dependencies from a predefined component knowledge base. It also needs to identify whether each component is a standard part, a parametric feature, or a custom complex shape, and set different parametric strategies for different types. Finally, parameter association and binding are performed, binding discrete structural parameters (e.g., length L=100, diameter Φ=20) with the specific components or assembly relationships between components identified in step one, forming a parameter-component association matrix. Then, the system automatically identifies engineering constraints between parameters (such as "hole diameter D must be less than wall thickness T" and "multiple bolt holes must be equidistant") and formalizes them into constraint rule expressions.

[0048] Furthermore, through the engineering design rule library of the CAD platform, the parameter association binding information and structural constraint relationship information are subjected to semantic rule matching processing by the relevant rule engine to obtain specific design instructions and derived design parameters.

[0049] In one embodiment, an extensible engineering design rule library can be built-in, with rules stored in the form of "IF<condition>THEN<action>". For example: "IF 'Design Goal' contains 'Lightweight' AND 'Component Type' = 'Support Bracket' THEN 'Initiate Topology Optimization Preprocessing', setting the material to aluminum alloy". The rule engine then takes the generated semantic information (design goal, component type, constraints) as input and triggers matching rules according to priority. The result of rule execution is the generation of a series of specific design instructions and derived design parameters.

[0050] Furthermore, through the CAD command-parameter mapping table, the specific design instructions and derived design parameters are mapped to the relevant CAD command parameter slots. Based on the constraint solver, the mapped constraint relationship parameter group is settled so that all parameter values ​​meet the product engineering constraints, resulting in the settled modeling process table.

[0051] In one implementation, a CAD command-parameter mapping table is configured, defining the parameter sets and their data types required for specific modeling commands in the CAD software (such as "Extrude" and "Drill"). For example, the "Extrude" command is mapped to the parameter set: {Sketch Profile, Extrude Depth, Draft Angle}. Then, the original structural parameters and new parameters derived from the rules are filled into the parameter slots of the corresponding CAD commands according to the mapping table. For parameter groups with constraints, a lightweight constraint solver (such as a numerical iteration-based method) is invoked to solve the constraints, ensuring that all parameter values ​​satisfy all engineering constraints.

[0052] Furthermore, the trained parametric modeling script engine generates a product parametric modeling script by processing the process content and design parameters in the modeling process table into command scripts. That is, based on the component relationship diagram, the sequential logic of modeling operations is planned according to the order of "basic parts → additional features → assembly relationships," generating a modeling process list. For example, a base plate must be created first before positioning holes can be drilled on it. Then, based on the process content and design parameters in the modeling process table, the corresponding product parametric modeling script is generated.

[0053] In one embodiment, a persistent model parameter storage module can also be used to persistently save component information and geometric parameters generated during each modeling or parametric modeling script generation process, supporting subsequent querying, modification, and version management.

[0054] S104. Perform 3D model building and execution processing on the product parametric modeling script based on key dimensions and key component types to generate a 3D product model.

[0055] Specifically, the product parametric modeling script is first decomposed into atomic and ordered components to obtain a modeling instruction sequence. Then, based on the parent-child relationships and geometric dependencies in the modeling instruction sequence, a feature dependency graph is generated. Next, according to the modeling instruction sequence and the feature dependency graph, virtual assembly constraints are automatically applied to each component of the product, and spatial conflict pre-resolution is performed to obtain the initial 3D model.

[0056] In one embodiment, an instruction parser is used to break down the script into an atomic, ordered sequence of modeling instructions (e.g., "create datum plane A", "draw sketch B on plane A", "extrude sketch B to form feature C"). Then, the parent-child relationships and geometric dependencies between instructions are analyzed to automatically construct a feature dependency graph. For example, it is identified that "bevel feature D" depends on "edge E", and "edge E" is generated by "extrude feature C", thus clarifying the sequential logic and data flow of modeling. Based on the modeling instruction sequence and the feature dependency graph, for products involving multiple parts, the execution engine maintains a virtual assembly manager. While building each part model, its key geometric reference elements (e.g., coordinate origin, datum axis, mating surfaces) are recorded synchronously. Once all relevant part models are generated in their respective documents, the engine automatically inserts the parts into a single assembly document based on the assembly relationships defined in the script (e.g., "concentric", "coincident", "distance"), and calls the CAD API to apply corresponding mating constraints, dynamically constructing the complete assembly. During or after the assembly constraints are applied, the engine automatically initiates a lightweight interference check to calculate whether there is volume overlap between parts. If a slight interference is detected (such as due to tolerance accumulation), the engine automatically correlates back to the driving parameters of the interfering part and performs micro-iterative adjustments within the tolerance range allowed by the parameters (such as slightly increasing a certain length parameter by 0.01mm). Then, the part is regenerated and the assembly is updated until the interference is eliminated.

[0057] Furthermore, by using the key semantic features corresponding to the user's original demand information, product component information, and product structural parameter information, the initial 3D model is verified based on the key dimensions and key component types to check the core structural parameters of the product. Based on the verification results, the product 3D model is obtained.

[0058] In one embodiment, by utilizing the key semantic features corresponding to the extracted user original demand information, product component information, and product structural parameter information, the key dimensions (such as total length, aperture, and assembly clearance) in the final model are automatically extracted, compared with the input product structural parameter information, and a dimensional conformity report is generated. If the deviation exceeds the threshold, a warning is issued. If the verification result is that the verification is passed, the product 3D model is automatically generated.

[0059] Furthermore, if the user's assessment of the product's 3D model indicates that the requirements are met, the product's 3D model is directly fed back to the user. If the user's assessment of the product's 3D model indicates that the requirements are not met, the newly added or deleted text is input into the user interface, and the 3D model construction process is re-executed until the user's assessment of the new product's 3D model indicates that the requirements are met.

[0060] As a feasible implementation method, if the user's judgment result of the product 3D model is that the requirements are not met, the model configuration items of the relevant component information and geometric parameters in the product 3D model are modified as needed, and the script regeneration and model reconstruction are automatically triggered to obtain the adjusted product 3D model.

[0061] In one embodiment, after the 3D model is displayed to the user, the user judges whether the model meets the requirements. If the user's judgment is that the requirements are met, the process ends; if not, feedback can be provided in two ways: Method 1: Re-enter or supplement the text requirements, and the system re-executes steps S101-S104 until the user's judgment is that the requirements are met; Method 2: Directly modify the component information or geometric parameters saved in the persistent storage module of the model parameters (such as adjusting the length of a component, changing the connection method, etc.). The system automatically triggers script regeneration and model reconstruction. When re-executing the modeling process, the system can reuse the stored component and parameter information and perform incremental updates in combination with the newly input requirements to improve response efficiency.

[0062] In addition, embodiments of this application also provide a hierarchical text-driven 3D modeling device, such as... Figure 2 As shown, the hierarchical text-driven 3D modeling device 200 specifically includes: At least one processor 201; and a memory 202 communicatively connected to the at least one processor 201; wherein the memory 202 stores instructions executable by the at least one processor 201 to enable the at least one processor 201 to execute: The requirement feature decomposition engine analyzes the received product design language requirement information to parse relevant functional components and component topology relationships, resulting in structured product component information. Based on the user's original requirements and product component information, the potential geometric parameters and parameters of potential related structural components in the specific geometry of each component of the product are deduced to obtain the product structure parameter information. By using the user's original requirements information, structured product component information, and product structure parameter information, the execution commands of the CAD platform are transformed into relevant process content and process design parameters to obtain the product parametric modeling script. The product parametric modeling script is processed to build and execute a 3D model based on key dimensions and key component types, generating a 3D model of the product.

[0063] This application's embodiments are centered on cloud-based 3D CAD architecture. By constructing a hierarchical parsing process, an editable parameter feedback mechanism, and a multi-level structured collaborative training system, it achieves efficient and controllable generation from fuzzy text descriptions to precise 3D models. Moreover, users do not need to master complex modeling commands or possess professional design experience; they can complete 3D model construction simply through natural language descriptions. This allows junior designers, non-technical personnel, and even ordinary users to participate in the product design process, promoting the widespread application of CAD technology. Furthermore, by combining LLM (Limited Language Management), "text instructions" are converted into "modeling scripts" that can be recognized and executed by the CAD system. Through a complete technical link of "text instructions - modeling scripts - CAD models," rapid generation of CAD models is achieved, significantly improving modeling efficiency. In addition, users can not only make overall adjustments through text feedback but also directly edit component structures and geometric parameters, achieving a closed-loop iteration of "semantic interpretability and parameter editability," balancing ease of use and accuracy.

[0064] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0065] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0066] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0067] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0068] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0069] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0070] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0071] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0072] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0073] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0074] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of this specification.

Claims

1. A hierarchical text-driven 3D modeling method, characterized in that, The method includes: The requirement feature decomposition engine analyzes the received product design language requirement information to parse relevant functional components and component topology relationships, resulting in structured product component information. Based on the user's original requirements and the product component information, the potential geometric parameters and parameters of potential related structural components in the specific geometry of each component of the product are deduced to obtain the product structure parameter information. Using the user's original requirements information, the structured product component information, and the product structure parameter information, the execution commands of the CAD platform are transformed into relevant process content and process design parameters to obtain the product parametric modeling script. The product parametric modeling script is processed to build a 3D model based on key dimensions and key component types, generating a 3D model of the product.

2. The hierarchical text-driven 3D modeling method according to claim 1, characterized in that, Before using a requirement feature decomposition engine to parse and process the received product design language requirement information for relevant functional components and component topology relationships to obtain structured product component information, the method further includes: Before the demand feature decomposition engine performs analysis, a text-driven 3D modeling task is constructed; wherein, the 3D modeling task has several training samples; Configure key fields in the JSON object of each training sample to obtain a fine-tuned subset of the 3D modeling task in different combinations of input and output; wherein, the key fields include: unique identifier, application domain, difficulty level, natural language modeling requirements input by the user, list of decomposed functional components, geometric parameters of each component, and executable parametric modeling script; The required feature decomposition engine, geometric parameter derivation engine, and parametric modeling script engine are trained using the fine-tuning subset based on temporal consistency features to obtain the trained component engines.

3. The hierarchical text-driven 3D modeling method according to claim 1, characterized in that, The requirement feature decomposition engine analyzes the received product design language requirements information, processing it to parse functional components and component topology relationships, resulting in structured product component information, specifically including: The system receives product design language requirements information input by the user in natural language form through a user interaction interface. Using a large language model, the product design language requirement information is processed through semantic grouping, structural independence, and text content parsing under state constraints to obtain preliminary product analysis information. Based on the trained requirement feature decomposition engine, the key structural features related to all requirements of the covering components in the preliminary product parsing information are decomposed into functional components to obtain the key structural features of the product. Based on the trained demand feature decomposition engine, the structural constraint decomposition of the component topology relationship is performed on the core spatial constraint features of the demand components and the core spatial constraint features containing other components in the preliminary product parsing information to obtain the product core spatial constraint features of the interrelated components. Based on the trained requirement feature decomposition engine and the core spatial constraint features of the product, the preliminary product parsing information is subjected to feature requirement decomposition processing to determine whether additional link structures are needed, thereby obtaining additional link structure features. The product's key structural features, core spatial constraint features, and additional link structural features are labeled as components of the product's basic structural units to obtain the product component information.

4. The hierarchical text-driven 3D modeling method according to claim 1, characterized in that, Based on the user's original requirements and the product component information, the potential geometric parameters and parameters of potential related structural components in the specific geometry of each component of the product are deduced to obtain the product structural parameter information, specifically including: Using a large language model, the user's original demand information is processed through semantic text parsing to obtain product demand and corresponding text information. The trained geometric parameter inference engine, based on the product requirements and corresponding text information, performs data recognition processing on the specific geometric parameters of each component in the product component information to obtain the identified intuitive structural parameter information; wherein, the specific geometric parameters include: geometric type parameters, size parameters, and three-dimensional spatial coordinates; Based on the intuitive structural parameter information, and through the geometric parameter deduction engine, the potential geometric parameters required for each component in the product component information and the potential related structural components are used to perform data deduction and simulation prediction under the relevant modeling requirements to obtain potential structural parameter information; wherein, the potential structural parameter information is structural parameter information that has not been directly identified. The intuitive structural parameter information and the potential structural parameter information are combined under the same product modeling parameters to generate the product structural parameter information.

5. The hierarchical text-driven 3D modeling method according to claim 1, characterized in that, Based on the user's original requirements information, the structured product component information, and the product structural parameter information, the execution commands of the CAD platform are transformed into relevant process content and process design parameters to obtain a product parametric modeling script, specifically including: Using a large language model, the user's original demand information, the product component information, and the product structure parameter information are processed to identify product component types and construct a component relationship graph, and the parameter association binding information and structural constraint relationship information of the product are extracted accordingly. By using the engineering design rule library of the CAD platform, the parameter association binding information and structural constraint relationship information are matched with the semantic rules of the rule engine to obtain specific design instructions and derived design parameters. The CAD command-parameter mapping table is used to map the specific design commands and derived design parameters to the relevant CAD command parameter slots. Based on the constraint solver, the mapped constraint relationship parameter group is calculated so that all parameter values ​​meet the product engineering constraints, resulting in the calculated modeling process table. The trained parametric modeling script engine generates the product parametric modeling script by processing the process content and process design parameters in the modeling process table into a command script.

6. The hierarchical text-driven 3D modeling method according to claim 1, characterized in that, The product parametric modeling script is processed to build a 3D model based on key dimensions and key component types, generating a 3D model of the product. Specifically, this includes: The product parametric modeling script is split into atomic and ordered components to obtain a modeling instruction sequence; Based on the parent-child relationships and geometric dependencies in the modeling instruction sequence, a feature dependency graph is generated. Based on the modeling instruction sequence and the feature dependency graph, the virtual assembly constraints of each component of the product are automatically applied and the parameters for spatial conflict are pre-resolved to obtain the initial three-dimensional model. By using the key semantic features corresponding to the user's original requirements information, the product component information, and the product structural parameter information, the initial 3D model is verified based on the key dimensions and key component types of the product's core structural parameters, and the product 3D model is obtained based on the verification results.

7. The hierarchical text-driven 3D modeling method according to claim 1, characterized in that, After performing 3D model building and execution processing on the product parametric modeling script based on key dimensions and key component types to generate a product 3D model, the method further includes: If the user's judgment result of the product 3D model is that the requirements are met, then the product 3D model will be directly fed back to the user. If the user's judgment result of the product's 3D model is that the requirements are not met, the newly added or deleted text will be input into the user interaction interface, and the 3D model construction process will be re-executed until the user's judgment result of the new product's 3D model is that the requirements are met.

8. The hierarchical text-driven 3D modeling method according to claim 7, characterized in that, If the user's judgment result of the product 3D model is that the requirements are not met, the model configuration items of the relevant component information and geometric parameters in the product 3D model will be modified as needed, and the script will be automatically triggered to regenerate and reconstruct the model to obtain the adjusted product 3D model.

9. A hierarchical text-driven 3D modeling device, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform a hierarchical text-driven 3D modeling method according to any one of claims 1-8.

10. A non-volatile computer storage medium, characterized in that, The storage medium is a non-volatile computer-readable storage medium that stores at least one program, each program including instructions that, when executed by a terminal, cause the terminal to perform a hierarchical text-driven 3D modeling method according to any one of claims 1-8.

Images