Optimization design method and system for mechanical part driven by large model

Abstract

The invention discloses an optimization design method of a large-model-driven mechanical part, which comprises the following steps of: performing knowledge retrieval according to a design task and a constraint condition, and generating text description of a structure design scheme by fusing the retrieved knowledge through a large-scale language model (LLM); the LLM performs task decomposition and modeling code mapping on the structure design scheme described by the text, and generated modeling codes are directly executed by modeling software to generate a three-dimensional model; the LLM generates a reasoning track and calls a finite element analysis tool set to perform performance analysis on the generated three-dimensional model; and according to optimization requirements and design constraints, the LLM optimizes the three-dimensional model to generate an optimal three-dimensional model. The invention discloses an optimization design system of a large model driven mechanical part. The optimization design system comprises a design scheme generation module, a parametric modeling module, a design scheme verification module and a structure parameter optimization module. The method and the system provided by the invention can save code quantity, reduce manual intervention and improve design efficiency.

Description

Technical Field

[0001] The present invention relates to the field of computer-aided technology, and in particular to a large-model driven optimization design method and system for mechanical components. Background Art

[0002] Mechanical structure design and optimization is a core part of the engineering field. Its fundamental task is to obtain the optimal structural solution under preset engineering constraints through systematic theoretical analysis, numerical simulation and experimental verification based on specific functional requirements and performance indicators (Pereira JLJ, Oliver GA, Francisco MB, et al. A Review of Multi-objective Optimization: Methods and Algorithms in Mechanical Engineering Problems[J]. Archives of Computational Methods in Engineering, 2022, 29(4): 2285-2308.). This complex process usually follows an established process covering multiple key stages such as requirements analysis, conceptual design, detailed design, performance analysis and solution verification. Traditional mechanical structure design methods are highly dependent on the designer's deep professional knowledge, rich engineering experience and proficiency in professional design and analysis software (Wang T, Wu D. Computer-Aided Traditional Art Design Based on Artificial Intelligence and Human-Computer Interaction[J]. Computer-Aided Design and Applications, 2023: 59-73.).

[0003] Take gas turbine blades as an example: Gas turbine blades are core components of gas turbines, and their design directly impacts the overall performance and reliability of the gas turbine. Traditional turbine blade design methods primarily rely on a combination of empirical design, computational fluid dynamics analysis, and finite element analysis. Based on experience and theoretical knowledge, designers initially determine the blade's aerodynamic shape, structure, and cooling scheme. They then use CFD and FEA software to evaluate its aerodynamic performance, stress distribution, and temperature field, repeatedly revising and optimizing the design based on simulation results. This approach suffers from limitations such as long design cycles, limited optimization space, and reliance on designer experience, making it difficult to fully realize the blade's performance potential.

[0004] Therefore, the current mechanical structure design and optimization have at least the following defects: (1) Low optimization efficiency: After the designer completes the preliminary CAD modeling of the turbine blade component, a series of tedious simulation pre-processing tasks such as meshing, boundary condition setting, and solution calculations are required by professional analysis engineers. After the analysis is completed, the results are fed back to the designer for modification. The entire "design-analysis-modification" closed-loop iteration process is time-consuming and has a significant impact on the product's speed to market.

[0005] (2) High reliance on personal experience: The quality of the initial concept plan depends largely on the knowledge level and project experience of the design engineer. For complex structures such as turbine blades, designers often refer to previous cases, making it difficult to conceive of innovative, high-performance topological configurations in the early stages. Design solutions often tend to be conservative and lack innovation.

[0006] (3) Strong result dependence: The quality of the optimization results depends largely on the experience level of the design engineer. Traditional optimization methods mostly rely on engineers to manually adjust parameters based on simulation results. They belong to local optimization and it is difficult to guarantee the global optimal solution. Usually, only a local optimal solution or suboptimal solution can be obtained.

[0007] (4) Difficulty in multi-objective optimization: In practical engineering problems, it is often necessary to consider multiple conflicting performance indicators simultaneously, for example, reducing weight as much as possible while ensuring stiffness. Traditional methods have difficulty in systematically weighing and optimizing multiple objectives. Summary of the Invention

[0008] The purpose of the present invention is to provide a large-model driven optimization design method and system for mechanical components. This method and system construct a closed-loop design and optimization workflow, which can save code volume, reduce manual intervention and improve design efficiency in the optimization design of mechanical components.

[0009] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions: The present invention provides a large model driven optimization design method for mechanical components, the method comprising the following steps: (1) Knowledge retrieval is performed based on the design tasks and constraints input by the designer in natural language form, and then the retrieved knowledge is integrated into the text description of the structural design scheme through the large language model (LLM); (2) The Large Language Model (LLM) decomposes the structural design scheme described in the text into tasks and maps it to modeling code. The generated modeling code is directly executed by the modeling software to generate a three-dimensional model. (3) Based on the predefined finite element analysis tool set, the large language model LLM generates an inference trajectory and calls the finite element analysis tool set to perform performance analysis on the generated three-dimensional model. If it does not meet the design task, step (4) is executed; (4) Based on the optimization requirements and design constraints proposed by the designer in natural language, the large language model (LLM) optimizes the three-dimensional model and finally feeds back the optimal three-dimensional model to the designer in a visual way.

[0010] In step (1), the retrieval augmentation generation (RAG) technology is used to index, retrieve and generate documents with high relevance to the design tasks and constraints from the professional knowledge base; context information is constructed based on the highly relevant documents, and is combined with the design tasks and constraints to form the final prompt word and input it into the large language model (LLM) to generate a text description of the structural design scheme.

[0011] In step (2), the method of performing task decomposition and modeling code mapping on the structural design scheme described in the text by the large language model LLM to generate a three-dimensional model includes: (2-1) The Large Language Model (LLM) decomposes the structural design plan into a series of subtasks, and gradually generates the CAD operation steps for each step and supplements the necessary spatial information, including shape, size, and positioning; (2-2) Large Language Model (LLM) converts a series of subtasks into parametric CAD modeling codes based on parametric CAD script library; (2-3) The generated parametric CAD modeling code is directly executed by the CAD software to generate a three-dimensional model.

[0012] Among them, if an error occurs during the execution of the CAD modeling code in steps (2-3), the CAD software will return error information to the LLM, and the LLM will make corrections based on the error information. This process will be iterated until executable CAD modeling code is generated.

[0013] In step (2), the large language model performs task decomposition and modeling code mapping on the structural design scheme described in the text to generate a three-dimensional model, including: (2-4) Interactive editing: Designers adjust the initially generated model by inputting new natural language instructions. After the adjustment, designers can preview the modification results in real time in the visual interface. Through multiple rounds of iterations, the initially generated 3D model can be dynamically modified and optimized to achieve interactive editing.

[0014] In step (3), the method of calling the finite element analysis toolset to perform performance analysis on the generated three-dimensional model includes: importing and preprocessing the three-dimensional model, defining material properties, defining and parameterizing key geometric features, setting physical fields and dividing the mesh, setting boundary conditions, solving and visualizing results.

[0015] The mechanical component is a gas turbine blade, a CNC machine tool column or an exchanger.

[0016] When the mechanical component is a gas turbine blade, the tasks in step (2) are decomposed into: wing-shaped contour generation, hollow airfoil model generation, internal spoiler rib generation, external film cooling hole generation and blade root structure generation; when the mechanical component is a CNC machine tool column, the tasks in step (2) are decomposed into: trapezoidal support column base generation, rectangular opening creation and Boolean operation, column solid shell extraction and internal rib arrangement.

[0017] The optimization design method provided by the present invention can solve the technical problems in the prior art of turbine blade component design, such as long process, low efficiency, disconnection between design and analysis, over-reliance on engineers' personal experience, lack of innovation and insufficient optimization.

[0018] The present invention also provides a large model driven optimization design system for mechanical components using the above method, the system comprising: The design solution generation module performs knowledge retrieval based on the design tasks and constraints input by the designer in natural language, and then integrates the retrieved knowledge into a text description of the structural design solution through the Large Language Model (LLM). The parametric modeling module, the Large Language Model (LLM), decomposes the structural design scheme described in the text into tasks and maps them to modeling code. The generated modeling code is directly executed by the modeling software to generate a 3D model. In the design solution verification module, the Large Language Model (LLM) generates reasoning trajectories and uses the finite element analysis tool set to perform performance analysis on the generated 3D model. The structural parameter optimization module uses the large language model (LLM) to optimize the 3D model based on the optimization requirements and design constraints proposed by the designer in natural language, and ultimately provides the optimal 3D model to the designer in a visual manner.

[0019] The specific implementation of each module in the system provided by the present invention refers to the above-mentioned optimization design method and will not be repeated here.

[0020] The present invention also provides a large model-driven optimization design device for mechanical components, comprising a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the above-mentioned large model-driven optimization design method for mechanical components.

[0021] The present invention also provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, is used to implement the above-mentioned large model-driven optimization design method for mechanical components.

[0022] Compared with the prior art, the present invention has the following excellent effects: (1) The present invention integrates four core functions: design solution generation, parametric modeling, design solution verification, and structural parameter optimization, and constructs a closed-loop automated design and optimization workflow, thereby achieving comprehensive automation and efficient connection of the design and optimization process, and effectively saving code volume; (2) The present invention implements a natural language-driven parametric modeling method and effective analysis of unstructured design intent in the parametric modeling process, supports an interactive design process, and significantly reduces the degree of dependence on the CAD skills of domain experts in the design process; (3) The present invention can reduce 50% to 80% of manual intervention in multiple design links (such as parametric modeling), thereby reducing labor costs and significantly improving overall design efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A process for generating an airfoil profile in a three-dimensional model for parametric modeling of a gas turbine blade in Example 1; Figure 2 A process for generating a hollow airfoil model in a three-dimensional model for parametric modeling of a gas turbine blade in Example 1; Figure 3 A process for generating an internal spoiler rib in a three-dimensional model for parametric modeling of a gas turbine blade in Example 1; Figure 4 The process of generating the airfoil profile in the three-dimensional model generated by parametric modeling of the gas turbine blade in Example 1; Figure 5 A process for generating a blade root structure in a three-dimensional model for parametric modeling of a gas turbine blade in Example 1; Figure 6 defining cooling channel geometric entities during the performance analysis of the three-dimensional model of the gas turbine blade generated in Example 1; Figure 7 Setting the physical field and dividing the mesh during the performance analysis of the three-dimensional model of the gas turbine blade generated in Example 1; Figure 8 To solve and visualize the performance analysis process of the three-dimensional model of the gas turbine blade generated in Example 1; Figure 9 Configuration A of the cooling pipe in the finite element analysis results of the turbine blade structure and surface temperature in Example 1; Figure 10 Configuration B of the cooling pipe in the finite element analysis results of the turbine blade structure and surface temperature in Example 1; Figure 11Configuration C of the cooling pipe in the finite element analysis results of the turbine blade structure and surface temperature in Example 1; Figure 12 Configuration D of the cooling pipe in the finite element analysis results of the turbine blade structure and surface temperature in Example 1; Figure 13 The comparison results of the three-dimensional model of the gas turbine blade generated in Example 1 under different optimization methods; Figure 14 The Pareto front diagram of the turbine blade optimization results in Example 1; Figure 15 is the initial design structure of the turbine blade in Example 1; Figure 16 This is a steady-state thermal analysis of the initial design structure of the turbine blade in Example 1; Figure 17 The optimal solution design structure of the turbine blade in Example 1 is as follows; Figure 18 Steady-state thermal analysis of the optimal solution design structure for the turbine blade in Example 1; Figure 19 The process of generating a three-dimensional model for the parametric modeling of the machine tool column in Example 2; Figure 20 Model import and attribute definition during the performance analysis of the three-dimensional model of the machine tool column generated in Example 2; Figure 21 Setting the physical field and dividing the mesh during the performance analysis of the three-dimensional model of the machine tool column generated in Example 2; Figure 22 Boundary conditions are set during the performance analysis of the three-dimensional model of the machine tool column generated in Example 2; Figure 23 To solve and visualize the results during the performance analysis of the three-dimensional model of the machine tool column generated in Example 2; Figure 24 The comparison results of the three-dimensional model of the machine tool column generated in Example 2 under different optimization methods. DETAILED DESCRIPTION

[0024] The large-model driven mechanical component optimization design method and system proposed in the present invention are further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] The present invention proposes a large-model driven optimization design method and system for mechanical components that integrates large-scale language models, CAD / CAE tools, and optimization algorithms for automated and intelligent design, including: The design solution generation module performs knowledge retrieval based on the design tasks and constraints input by the designer in natural language, and then integrates the retrieved knowledge into a text description of the structural design solution through the Large Language Model (LLM). The parametric modeling module, the Large Language Model (LLM), decomposes the structural design scheme described in the text into tasks and maps them to modeling code. The generated modeling code is directly executed by the modeling software to generate a 3D model. In the design solution verification module, the Large Language Model (LLM) generates reasoning trajectories and uses the finite element analysis tool set to perform performance analysis on the generated 3D model. The structural parameter optimization module uses the large language model (LLM) to optimize the 3D model based on the optimization requirements and design constraints proposed by the designer in natural language, and ultimately provides the optimal 3D model to the designer in a visual manner.

[0026] Example 1 Design Optimization of Gas Turbine Blades The method for optimizing the design of gas turbine blades using the large model-driven mechanical component optimization design system provided by the present invention comprises the following steps: Step 1: Designers ask questions and describe requirements Designers use natural language to input high-level design tasks and constraints into the design solution generation module. In this example, the task is specifically, "How do I design the structure of a gas turbine blade?" Or, more specifically, "Design a high-temperature-resistant, lightweight gas turbine blade with an inlet temperature of 1100K. The average surface temperature of the blade after cooling must be below 840K, and the overall mass must be less than 10.80 kg."

[0027] Step 2: Knowledge Retrieval and Recall The design scheme generation module uses RAG technology to retrieve knowledge related to keywords such as "gas turbine blade shaping" and "gas turbine turbine cooling blade design and optimization" from the internal knowledge base (including historical design cases, technical patents, academic papers, design specifications, etc.); the design scheme generation module (based on LLM) integrates the retrieved knowledge to generate a preliminary design scheme corresponding to the design requirements of the designer.

[0028] In this embodiment, the knowledge base retrieval results (highly relevant documents) provide the design direction and key constraints for the design scheme generation module: Document 1 explains the importance of high-temperature nickel-based alloy materials in turbine blade design, emphasizing their strength, creep resistance, and oxidation resistance; Document 2 explains the diversity of turbine blade airfoil selection and the key role of airfoil design in optimizing aerodynamic performance; Document 3 explains the turbine blade cooling strategy, emphasizing the role of internal cooling channels and external film cooling in improving cooling efficiency.

[0029] The core mechanism of RAG is to organically integrate information retrieval and text generation, so that LLM can dynamically obtain relevant information from external knowledge bases during the generation process to make up for the shortcomings of its own knowledge. The workflow of RAG includes three main steps: indexing, retrieval and generation. The foundation of RAG is to build a professional knowledge base and a design resource base, and to perform efficient knowledge retrieval based on this. The main data sources of the professional knowledge base in the present invention include "Mechanical Design Manual", "Mechanical Principles", "Gas Turbine Blade Modeling", "Gas Turbine Turbine Cooling Blade Design and Optimization", etc. The design resource base mainly includes existing gas turbine turbine cooling blade design and optimization cases, which include design tasks of various complexities. The case library provides reference and reference for the generation of design solutions, which helps LLM to generate more innovative and practical solutions.

[0030] Step 3: Design solution generation Through the retrieved highly relevant documents, the design solution generation module constructs contextual information and combines it with the designer's original question to form the final prompt word, which is input into the LLM to generate a preliminary design solution.

[0031] In this example, the generated structural design solution describes a high-temperature blade design using GTD-111 directionally solidified nickel-based alloy. A parameterized NACA four-axis airfoil is selected, with multi-channel serpentine cooling ducts and multi-directional spoiler ribs used internally to enhance heat transfer. Externally, cylindrical cooling holes provide air film protection. Thermal simulation analysis and iterative structural parameter iterations balance aerodynamic performance, cooling efficiency, and structural strength, ensuring safe and reliable operation of the blade at high temperatures.

[0032] Step 4: Parametric modeling task decomposition Use parametric modeling modules to transform design plans into precise CAD models. Specifically: The parametric modeling module first simulates the "divide and conquer" strategy of human designers through LLM, breaking down complex tasks into a series of more manageable subtasks, gradually generating CAD operation steps for each step and supplementing the necessary spatial information. The process can be expressed as follows: ; in, Indicates the preliminary design plan. represents the total number of design steps, Indicates the i Description of the design steps, represents the probability distribution of the design steps generated by LLM under given conditions, Represents a given example.

[0033] It not only includes specific operation content, but also details key spatial geometric information such as shape, size, and positioning to ensure the correctness and robustness of subsequent CAD code generation.

[0034] In the process of generating the CAD model of the gas turbine turbine blade in this embodiment, a parametric modeling module is used to decompose it into five steps, specifically: wing-shaped contour generation -> hollow airfoil model generation -> internal spoiler rib generation -> external film cooling hole generation -> blade root structure generation.

[0035] Step 5: Parametric Modeling Code Mapping The parametric modeling module leverages the semantic understanding and code generation capabilities of the LLM to convert natural language descriptions into parametric CAD modeling code based on the CadQuery library. The CadQuery library, with its concise syntax and powerful parametric capabilities, is an ideal choice for generating 3D models using the method described in this paper. This process can be formally expressed as: ; in, Represents the initially generated CAD script code, represents the conditional probability distribution of LLM generating feasible CAD scripts under given input conditions, Indicates the specific steps of the modeling task, such as positioning information, shape and size, stretching and rotation, etc. Generated CAD code Directly executed by CAD software, calling the modeling engine to generate a 3D model M Designers can preview the model’s geometry and parameter settings in real time in the visual interface to assess whether it meets the design requirements. This process can be expressed as: ; If an error occurs during the execution of the CAD code, the CAD software will return an error message to the LLM. , LLM will make corrections based on the error information, and this process will be iterated until executable code is generated. The process can be expressed as: ; In this embodiment, for a gas turbine blade, the generation process is as follows: (1) Wing shape contour generation: The parametric modeling module generates the two-dimensional contour of the NACA 4412 airfoil. Based on the pre-set parameter range, key parameters such as maximum camber, maximum camber position and maximum thickness are determined, and the coordinates of the airfoil contour points are calculated using the NACA 4-digit airfoil formula, such as Figure 1 As shown; (2) Hollow airfoil model generation: After obtaining the accurate airfoil profile, the parametric modeling module creates a three-dimensional solid through stretching operation as the outer wall of the hollow airfoil, and generates a smaller airfoil solid as the inner wall of the hollow structure according to a certain scaling ratio. The parametric modeling module then performs a Boolean difference operation to subtract the inner airfoil from the outer airfoil to obtain the hollow airfoil structure, as shown in Figure 2. Figure 2 As shown; (3) Generation of internal spoiler ribs: In order to improve the internal cooling effect, the parametric modeling module generates spoiler ribs inside the hollow airfoil, such as Figure 3 As shown; (4) External film cooling hole generation: The parametric modeling module generates the geometric parameters, positions and arrangements of the film holes, and creates the film holes through Boolean subtraction operations, such as Figure 4 As shown; (5) Blade root structure generation: The parametric modeling module generates the blade root outline by scaling the blade outline, and creates the upper and lower blade roots by lofting, such as Figure 5 As shown; Table 1 provides a quantitative evaluation of the efficiency of the parametric modeling module in the parametric modeling process. The data shows that the average one-time code generation success rate for each subtask reached 94%, indicating that the parametric modeling module can efficiently convert design intent into executable CadQuery code, and in most cases no manual correction is required. Specifically, in the steps of airfoil contour generation, hollow airfoil creation, and blade root structure generation, the one-time completion rate reached 100%. Even in more complex tasks such as internal spoiler ribs and film cooling hole generation, the one-time completion rate reached 90% and 80%, respectively. More importantly, compared with traditional manual parametric script writing, the code saving rate reached an average of 73.36%, which significantly reduced the modeling workload and improved the efficiency of design iteration. These results verify the significant advantages of the parametric modeling module in automating the parametric modeling process.

[0036] Table 1 Evaluation of the efficiency of turbine blade parametric modeling .

[0037] Step 6: Interactive Editing In the practical application of CAD model design, designers often find it difficult to obtain a model that fully meets their expectations by simply inputting commands. To address this problem, the present invention introduces an interactive editing module. This module allows designers to adjust the model by inputting new natural language commands, adjusting the model's geometric parameters, topology, and other aspects, thereby achieving dynamic modification and optimization of the model.

[0038] The interactive editing module achieves fine-tuning of the CAD model through multiple rounds of iteration. Specifically, given an initially generated 3D model, the designer can modify it by inputting new natural language instructions. The large language model generates updated CAD code based on the new instructions and the model status. The CAD software regenerates the model, and the designer views the modified results in a visual interface. If the model still does not meet expectations, the designer can continue to input new natural language instructions to make adjustments. This process can be expressed as: ; in, For the k The CAD script code generated by the iteration, New natural language instructions for designer input, The CAD script code generated for the previous iteration, For error feedback information, M ′ is the regenerated three-dimensional model.

[0039] The interactive editing module can simultaneously generate multiple design solutions with different topologies and present them to the designer, who can provide feedback and make edits using natural language.

[0040] Step 7: Design Verification To achieve efficient collaboration between LLM and finite element tools, this paper modularizes and encapsulates the finite element analysis process, converts discrete FEA operation steps into callable tool functions, and constructs a modular FEA toolset for the design solution verification module. As shown in Table 2, this toolset covers the entire process from model import, physical field configuration, to result analysis. Through standardized interfaces and functional encapsulation, these tools enable LLM to execute corresponding interface functions to help designers efficiently complete complex tasks such as geometric preprocessing, physical field settings, material property definition, and mesh optimization, significantly improving the automation level and scalability of the simulation process.

[0041] Table 2 FEA toolset .

[0042] The designer submits analysis requirements to the design scheme verification module in the form of natural language. Due to the high temperature of the blade simulation analysis, the designer's original question is: "Perform a steady-state thermal analysis of this turbine blade model, considering an inlet temperature of 1100K, a cooling gas temperature of 600K, a material of DS GTD111 nickel-based high-temperature alloy, an inlet pressure of 40 bar, a cooling gas pressure of 30 bar, and an inlet velocity of 450m / s. Please complete the finite element analysis step by step and tell me the minimum and average surface temperatures." To verify the cooling effect of the designed blade, the design scheme verification module generates an inference trajectory based on the predefined finite element analysis tool set, selects and executes the corresponding tool, and completes the analysis task through multiple "inference-action-observation" cycles. The specific implementation process is as follows: (1) Model import and preprocessing: The design scheme verification module calls the model import tool to load the STEP file generated by parametric modeling, automatically identifies the geometric entities of the turbine blades and establishes topological relationships; (2) Define material properties: The design scheme verification module calls the material setting tool to define the material properties of the DSGTD111 nickel-based high-temperature alloy for the turbine blades and the corresponding fluid properties for the cooling gas; (3) Define the cooling channel geometric entities: The goal of this step is to select the inner wall surface of the cooling channel. Since the LLM lacks the ability to accurately perceive spatial geometric information, it is difficult for the design verification module to automatically distinguish the inner wall of the cooling channel from other surfaces through the program, and manual assistance is required for boundary selection. The design verification module calls the manual assistance tool to present the geometric model to the designer, who manually selects the set of geometric entities corresponding to the inner wall of the cooling channel. (4) Setting the physical field and dividing the mesh: The design scheme verification module calls the physical field setting tool to initialize the physical field settings for steady-state thermal analysis, calls the mesh division setting tool, sets the mesh unit type to tetrahedral units, and divides the mesh according to the preset mesh density parameters; (5) Boundary condition setting: In this step, the design scheme verification module is responsible for applying various boundary conditions to simulate actual working conditions. However, due to the complexity of the geometric structure and the superposition of multiple boundary conditions, this step still requires manual assistance. The design scheme verification module calls the manual assistance tool, and the designer manually completes the selection of complex geometric areas and the boundary application operation. (6) Solution: The design scheme verification module calls the solution tool, starts the finite element solver, and performs steady-state thermal analysis calculations; (7) Result visualization: The design scheme verification module calls the result visualization tool to generate surface temperature results and feeds back the results to the designer: "I have completed the thermal finite element analysis of the turbine blade according to your description. The minimum temperature is 600.18 K and the average temperature is 885.13 K."

[0043] Among them, Figure 5 The model import and attribute definition shown in the figure define the cooling channel geometry entity as follows Figure 6 As shown, set the physical field and divide the mesh as Figure 7 As shown, the solution and result visualization are as follows Figure 8 As shown.

[0044] Table 3 presents the performance indicators of the design solution verification module during the FEA verification and structural optimization stages. In various subtasks of FEA verification, such as model import and preprocessing, material definition, and physical field setup, the one-time code / API call success rate reached 95% and 90%, respectively, and the corresponding code / operation savings rate was as high as approximately 80% and 70%. For the more challenging step of applying boundary conditions, especially when it comes to the selection of complex geometric features, the one-time success rate was approximately 60%, but LLM was still able to generate most of the framework code, contributing approximately 50% of the code / operation savings. Overall, the present invention's contribution to the automation of FEA tasks (based on the proportion of code / operation sequences) reached approximately 70%.

[0045] Table 3 Performance indicators of the design scheme verification module in the turbine blade FEA verification and optimization stage .

[0046] Step 8: Structural Optimization Based on parametric modeling and steady-state thermal analysis, in order to further improve the cooling efficiency of turbine blades, the structural parameters and configuration of the blades are optimized through the structural parameter optimization module, aiming to obtain design parameters with optimal cooling performance under high-temperature conditions.

[0047] The designers first modified the turbine blade model and defined four different cooling channel configurations. They then submitted optimization analysis requirements to the structural parameter optimization module in the form of natural language: "Perform structural optimization for the four configurations and corresponding parameters of the current turbine blade design scheme, and reduce the average temperature of the blade surface as much as possible while satisfying the constraint that the mass does not exceed 10.80 kg."

[0048] The structural parameter optimization module generates the objective function, constraints and optimized parameter range required by the optimization algorithm, calls optimization methods such as SAASBO, LHSBO, MGPBO, ALEBO and CoCaBO to optimize the structural parameters of turbine blades, and outputs feedback to designers.

[0049] The design space of turbine stator blades defined by the structural parameter optimization module is as follows: Figures 9-12As shown in the figure, the cooling channel configurations are named Configuration A, Configuration B, Configuration C, and Configuration D. Configuration A uses five basic internal spoiler ribs to form a U-shaped cooling channel and serves as the baseline solution. Configurations B and D add additional spoiler ribs inside the blade to increase surface contact with the cooling medium. Configurations C and D arrange cooling holes on the blade surface to enhance the cooling effect by forming a cooling film.

[0050] Table 4 defines all adjustable dimensions of the parameterized model of the turbine stator blade.

[0051] Table 4 Design space of turbine stator blades .

[0052] The structural parameter optimization module can formally define the optimization problem as: ; in, It is a continuous design variable that reflects the size and direction of the cooling channel; is a categorical variable corresponding to the configuration of the cooling channel. M(X) ≤10.80kg ensures that the blades achieve high-performance cooling while meeting lightweight design requirements.

[0053] like Figure 13 As shown in the figure, the structural parameter optimization module calls the optimization methods such as SAASBO, LHSBO, MGPBO, ALEBO and CoCaBO. After optimization, the average surface temperature of the cooling channel of the optimal solution is reduced to 840K, while meeting the quality constraints. M(X) Under the premise of ≤10.80kg, the cooling efficiency is significantly improved.

[0054] like Figure 14 As shown, the horizontal axis is the average temperature of the cooling pipe, and the vertical axis is the total mass of the turbine blade. M(X) Under the assumption that the mass is ≤10.80 kg, the optimization goal is to minimize temperature. Points above the dashed line exceed the mass constraint. Points connected by the solid green line are Pareto optimal solutions. The black points are the final design solutions selected by the designers from among all the optimal green points, based on their final trade-offs and preferences. The structural parameter optimization module applies the optimization algorithm to the four blade configurations proposed by the designers, A, B, C, and D. After optimization, the optimal solution is configuration D, with an average cooling channel temperature of 836.77 K and a total mass of 10.71 kg.

[0055] Comparison of structural and steady-state thermal analysis results between the initial design and the optimal solution design of the turbine blade case Figures 15-18As shown, the optimal solution reduces the average blade surface temperature by 23K and significantly improves blade cooling efficiency while meeting mass constraints. Experimental results demonstrate that the proposed method can effectively improve turbine blade cooling performance, providing a new technical approach for the development of high-performance gas turbines.

[0056] Combining the above steps, the method and system provided by the present invention have shown significant application potential in the case of gas turbine blade design and optimization: in the design scheme generation stage, the introduction of the RAG mechanism effectively improves the engineering feasibility of the preliminary concept; in the parametric modeling stage, through natural language-driven code generation, not only the requirements for professional CAD programming skills are greatly reduced, but also more than 85% of the modeling work is automated, and the average code volume is saved by more than 70%; in the design verification and optimization stage, through intelligent tool calling and process orchestration, about 70% of the FEA tasks are automated, and the optimization process is fully automatic. These quantitative results fully demonstrate the significant advantages of the method and system provided by the present invention in improving design efficiency and lowering technical barriers, especially in the intelligent coordination of automated code generation and complex processes, which has brought innovative changes to traditional mechanical design and optimization methods.

[0057] Example 2 Design Optimization of Five-Axis CNC Machine Tool Column Step 1: Designers ask questions and describe requirements Designers input high-level design tasks and constraints into the design solution generation module using natural language. In this example, the task is: "How can I design a high-rigidity, lightweight column for a five-axis CNC machine tool?" Or, more specifically, "Design a column for a five-axis CNC machine tool with a mass not exceeding 10 tons and a maximum deformation of less than 25 microns under an axial load of 2 kN." Step 2: Knowledge Retrieval and Recall The design solution generation module uses RAG technology to retrieve relevant knowledge related to keywords such as "machine tool column structure," "structural stiffness optimization," and "lightweight design" from an internal knowledge base (including historical design cases, technical patents, academic papers, and design specifications). The design generation module (based on the LLM) integrates this retrieved knowledge to generate a preliminary design solution that meets the designer's design requirements.

[0058] In this embodiment, the knowledge base search results provide the design direction and key constraints for the design scheme generation module, including the following highly relevant documents: Document 1 describes the design features and advantages of the gantry machine tool column structure; Document 2 describes how to improve the stiffness performance of the column by optimizing the arrangement of the ribs, emphasizing the key role of the ribs in bearing bending and torsional loads; Document 3 describes how to achieve performance optimization by adjusting the key structural parameters of the column.

[0059] Step 3: Design solution generation Through the retrieved highly relevant documents, the design solution generation module constructs contextual information and combines it with the designer's original question to form the final prompt word, which is input into the LLM to generate a preliminary design solution.

[0060] In this example, the generated structural design text describes the following: A classic dual-column gantry structure is employed, with the two columns rigidly connected by a top crossbeam to form a stable frame structure. The column cross-section is designed to be trapezoidal to accommodate load distribution and improve bending resistance. A rectangular internal cavity is introduced in the center of the column to form the left and right support columns, the top crossbeam, and the lower bed, significantly reducing weight while maintaining rigidity. Multiple layers of transverse and longitudinal ribs are installed within the column to improve the structure's bending and torsional rigidity. HT250 gray cast iron was initially selected as the column material, and subsequent simulation analysis and optimization iterations will further adjust the rib layout, wall thickness, and material selection to achieve the optimal balance between rigidity and lightweight.

[0061] Step 4: Parametric Modeling Code Mapping The parametric modeling module is used to convert the design plan into an accurate CAD model. The intermediate stage models and key operations in the modeling process are as follows: Figure 19 The specific steps are as follows: (1) Generation of the trapezoidal support column base: The parametric modeling module first generates a cross-sectional sketch of the trapezoidal main body base, and then performs biaxial stretching along the X-axis to form an initial three-dimensional solid; (2) Rectangular opening creation and Boolean operation: The parametric modeling module creates a rectangular cross-sectional profile in the middle of the trapezoidal support column base and stretches it along the Z-axis to form a solid body that penetrates the column. Subsequently, the parametric modeling module performs a Boolean difference operation to subtract the rectangular solid body from the column body, thereby dividing the original trapezoidal cross-sectional body into two symmetrical support columns on the left and right, as well as a top beam and a bottom base. (3) Shelling of column solids: To further reduce weight and reserve space for the rational arrangement of internal ribs, the parametric modeling module performs shelling operations on the column solid model, transforming it into a thin-walled structure with a specific wall thickness. The shelling thickness is controlled by parameters. (4) Internal rib arrangement: In order to improve the stiffness and stability of the column, the parametric modeling module creates a two-dimensional sketch of the rib at a specified position inside the column, then forms a three-dimensional entity through parametric stretching operations, and finally performs a Boolean union operation on the rib entity and the column body to combine them into a whole.

[0062] Step 5: Design Verification In order to evaluate the mechanical properties of the initial machine tool column model under actual working conditions, the design scheme verification module is used to perform static analysis on the column. The designer puts forward the analysis requirements to the design scheme verification module in the form of natural language: "Set the static analysis module of the machine tool column, the working load is a 3 kN load in the Z direction in front of the beam, and a fixed support is set on the bottom surface. The solution result is stress and strain." The design scheme verification module is based on the predefined finite element analysis (FEA) tool set and completes the analysis task through multiple "reasoning-action-observation" cycles. Figure 20-23 The specific implementation process is as follows: (1) Model import and preprocessing: The design scheme verification module calls the model import tool to load the STEP file generated by parametric modeling, automatically identifies geometric entities and establishes topological relationships; (2) Define material properties: The design scheme verification module calls the material setting tool to match the material properties of HT250 cast iron (elastic modulus 210 GPa, Poisson's ratio 0.28) for the column model; (3) Setting the physical field and dividing the mesh: The design scheme verification module calls the physical field setting tool to initialize the static analysis module, and calls the mesh division setting tool to set the mesh type to tetrahedral mesh; (4) Boundary condition setting: The design scheme verification module calls the named entity selection tool to create an entity set in the bottom surface area, and calls the boundary condition setting tool to impose fixed constraints on the bottom surface area; (5) Applying load: The design scheme verification module calls the named entity selection tool to create an entity definition for the front surface area of ​​the beam, and calls the load setting tool to apply a uniformly distributed load of 3 kN in the Z direction downward to this area; (6) Solving: The design scheme verification module calls the solving tool, starts the finite element solver, and performs static calculations; (7) Result visualization: The design scheme verification module calls the result visualization tool to generate visualization results of the global deformation and feeds back the results to the designer: "I have completed the static analysis of the machine tool column according to your description, and the global maximum deformation is 30.79 μm"; Among them, model import and attribute definition are as follows Figure 20 As shown, set the physical field and divide the mesh as Figure 21 As shown, the boundary conditions are set as Figure 22 As shown, the solution and result visualization are as follows Figure 23 shown.

[0063] Step 6: Structural Optimization Based on parametric modeling and static analysis, in order to further improve the stiffness of the machine tool column, the key structural dimensions of the column are optimized through the structural parameter optimization module. The design space of the machine tool column is shown in Table 5.

[0064] Designers submit optimization analysis requirements to the structural parameter optimization module in the form of natural language: "Optimize the structure and parameters of the current machine tool column design scheme, and minimize the global maximum deformation while satisfying the constraint that the mass does not exceed 1.70 tons." The structural parameter optimization module generates the objective function, constraints, and optimized parameter range required by the optimization algorithm, calls the optimization algorithm, and optimizes the structural parameters of the column.

[0065] The mathematical model for machine column optimization defined by the structural parameter optimization module is as follows: ; in It is a continuous design variable, reflecting the key dimensional parameters of the machine tool column; is a categorical variable corresponding to the material selected for the machine column; quality constraint Ensure lightweight design while meeting high rigidity requirements.

[0066] The structural parameter optimization module uses SAASBO, LHSBO, MGPBO, ALEBO and CoCaBO optimization methods for optimization. The comparison results of different optimization methods are shown in the figure. Figure 24 As shown in the figure, after optimization, the global maximum deformation of the optimal solution is reduced to 24.14 μm, which significantly improves the stiffness of the column while meeting the quality constraints.

[0067] Table 5 defines all adjustable dimensions of the parametric model of the machine column.

[0068] Table 5 Design space of machine tool column .

[0069] In summary, the optimization design method and system provided by the present invention also show significant application potential in the design and optimization case of five-axis CNC machine tool columns: in the design scheme generation stage, the RAG mechanism helps to quickly form preliminary concepts based on domain knowledge; in the parametric modeling stage, through natural language-driven code generation, it is expected to achieve a high modeling automation rate and code saving; in the design verification and optimization stage, through intelligent tool calling and process orchestration, partial automation of the FEA process and full automatic execution of the optimization process are achieved, and optimization results that are better than the baseline algorithm are achieved.

Claims

1. A large model driven optimization design method for mechanical components, characterized in that: The method comprises the following steps: (1) Knowledge retrieval is performed based on the design tasks and constraints input by the designer in natural language form, and then the retrieved knowledge is integrated into the text description of the structural design scheme through the large language model (LLM); (2) The Large Language Model (LLM) decomposes the structural design scheme described in the text into tasks and maps it to modeling code. The generated modeling code is directly executed by the modeling software to generate a three-dimensional model. (3) Based on the predefined finite element analysis tool set, the large language model LLM generates an inference trajectory and calls the finite element analysis tool set to perform performance analysis on the generated three-dimensional model. If it does not meet the design task, step (4) is executed; (4) Based on the optimization requirements and design constraints proposed by the designer in natural language, the large language model (LLM) optimizes the three-dimensional model and finally feeds back the optimal three-dimensional model to the designer in a visual way.

2. The large model driven optimization design method for mechanical components according to claim 1, characterized in that: In step (1), the retrieval augmentation generation (RAG) technology is used to index, retrieve and generate documents with high relevance to the design tasks and constraints from the professional knowledge base; context information is constructed based on the highly relevant documents, and is combined with the design tasks and constraints to form the final prompt word and input it into the large language model (LLM) to generate a text description of the structural design scheme.

3. The large model driven optimization design method for mechanical components according to claim 1, characterized in that: In step (2), the method of performing task decomposition and modeling code mapping on the structural design scheme described in the text by the large language model LLM to generate a three-dimensional model includes: (2-1) The Large Language Model (LLM) decomposes the structural design plan into a series of subtasks, and gradually generates the CAD operation steps for each step and supplements the necessary spatial information, including shape, size, and positioning; (2-2) Large Language Model (LLM) converts a series of subtasks into parametric CAD modeling codes based on parametric CAD script library; (2-3) The generated parametric CAD modeling code is directly executed by the CAD software to generate a three-dimensional model.

4. The large model driven optimization design method for mechanical components according to claim 3, characterized in that: In step (2), the large language model performs task decomposition and modeling code mapping on the structural design scheme described in the text to generate a three-dimensional model, including: (2-4) Interactive editing: Designers adjust the initially generated model by inputting new natural language instructions. After the adjustment, designers can preview the modification results in real time in the visual interface. Through multiple rounds of iterations, the initially generated 3D model can be dynamically modified and optimized to achieve interactive editing.

5. The large model driven optimization design method for mechanical components according to claim 1, characterized in that: In step (3), the method of calling the finite element analysis toolset to perform performance analysis on the generated three-dimensional model includes: importing and preprocessing the three-dimensional model, defining material properties, defining and parameterizing key geometric features, setting physical fields and dividing the mesh, setting boundary conditions, solving and visualizing results.

6. The large model driven optimization design method for mechanical components according to claim 1, characterized in that: The mechanical component is a gas turbine blade, a CNC machine tool column or an exchanger; when the mechanical component is a gas turbine blade, the task in step (2) is decomposed into: wing-shaped contour generation, hollow airfoil model generation, internal spoiler rib generation, external film cooling hole generation and blade root structure generation; when the mechanical component is a CNC machine tool column, the task in step (2) is decomposed into: trapezoidal support column base generation, rectangular opening creation and Boolean operation, column solid shell extraction and internal rib arrangement.

7. The large model driven optimization design method for mechanical components according to claim 1, characterized in that: In step (4), the large language model (LLM) generates the required standard functions, constraints and optimized parameter ranges for optimizing the three-dimensional model, and optimizes the three-dimensional model by calling the optimization algorithm.

8. A large model driven optimization design system for mechanical parts using the method according to any one of claims 1 to 7, characterized in that: The system comprises: The design solution generation module performs knowledge retrieval based on the design tasks and constraints input by the designer in natural language, and then integrates the retrieved knowledge into a text description of the structural design solution through the Large Language Model (LLM). The parametric modeling module, the Large Language Model (LLM), decomposes the structural design scheme described in the text into tasks and maps them to modeling code. The generated modeling code is directly executed by the modeling software to generate a 3D model. In the design solution verification module, the Large Language Model (LLM) generates reasoning trajectories and uses the finite element analysis tool set to perform performance analysis on the generated 3D model. The structural parameter optimization module uses the large language model (LLM) to optimize the 3D model based on the optimization requirements and design constraints proposed by the designer in natural language, and ultimately provides the optimal 3D model to the designer in a visual manner.

9. A large model driven optimization design device for mechanical parts, characterized in that: It comprises a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement the large model-driven optimization design method for mechanical components described in any one of claims 1-7.

10. A computer-readable storage medium having a program stored thereon, which, when executed by a processor, is used to implement the large model-driven optimization design method for mechanical components according to any one of claims 1 to 7.

Images