Natural language computer-aided engineering and computer-aided design method

Abstract

The invention discloses natural language computer-aided engineering and a computer-aided design method. CAE simulation analysis and design optimization on complex product design are high in threshold and difficulty. According to the method, an artificial intelligence large language model is finely adjusted by using professional knowledge of CAE and CAD, an agent is developed to call an API service of the large language model, an engineer dialogues with the agent by using a natural language, and the large language model can understand purposes and requirements of the engineer and generate corresponding operation instructions or program codes of CAE software and CAD software; the agent drives the CAE software and the CAD software to execute instructions or run programs to complete simulation analysis and design optimization tasks, and an engineer can cooperate with a plurality of professional CAE software and CAD software to perform simulation analysis and design optimization of multi-physics coupling only by dialogue with the agent. The intelligent agent can autonomously complete loop iteration simulation analysis and design optimization tasks according to formulated scheme steps and rules.

Description

Technical Field

[0001] The present invention relates to computer-aided engineering and computer-aided design, and in particular to a natural language computer-aided engineering and computer-aided design method. Background Art

[0002] Computer-aided engineering (CAE) is a technology that uses computer software to perform engineering simulation analysis to optimize product design and performance. CAE is widely used in various industrial fields to help engineers predict performance and solve potential problems in the early stages of product development, thereby reducing the need for physical testing, reducing costs and shortening development cycles. Computer-aided design (CAD) uses computer software for design and drawing, mainly for creating accurate two-dimensional or three-dimensional models. CAE software and CAD software are indispensable simulation analysis and design optimization tools in modern design and engineering. A basic input of CAE is the model of CAD design, and the simulation analysis results of CAE are used to guide CAD to improve and optimize the design.

[0003] As modern product design becomes more and more complex, the requirements for accuracy, performance and reliability are constantly increasing. The functions and integration of CAE and CAD software are also continuously improving, and the threshold for using the software is getting higher and higher. In particular, CAE software, because it involves complex physical models, numerical calculations and multidisciplinary coupling analysis, places higher demands on the user's theoretical knowledge, engineering experience and software operation capabilities. CAE software usually requires users to have a solid foundation in disciplines such as mechanics, thermodynamics, fluid dynamics, and electromagnetism. At the same time, they must also be familiar with the principles and application scenarios of numerical methods such as finite element analysis (FEA), computational fluid dynamics (CFD), and electromagnetic fields. In addition, as the simulation accuracy increases, the model's meshing, boundary condition setting, material property definition and other steps have become more complicated, and the slightest carelessness may lead to distortion of simulation results or calculation failure.

[0004] In order to lower the threshold for use, software developers are actively introducing intelligent technologies and automation functions. For example, through machine learning algorithms, CAE software can automatically optimize meshing, recommend appropriate solver settings, and even predict simulation results based on historical data. In addition, some software also provides a wizard-style operation interface and built-in templates to help beginners get started quickly. Despite this, the high threshold of CAE software still poses a challenge to companies and engineers. Companies need to invest a lot of resources in talent training and team building, while engineers need to constantly learn new knowledge and master new tools to keep up with the pace of technological development. There is still a big gap in the training of high-end simulation talents in my country.

[0005] In recent years, artificial intelligence large language models have made rapid progress. These models are trained based on massive data and can capture the complex patterns and contextual relationships of language, thus performing well in tasks such as text generation, translation, question answering, and code writing, greatly improving work efficiency. This also provides conditions for the development of CAE and CAD technologies towards intelligence, automation, and collaboration.

[0006] At present, there are text-to-CAD products based on machine learning models. They need to call a machine learning model API, which can generate CAD models based on text prompts and can be imported into the CAD software used by users. Users can also fine-tune this machine learning model with their own proprietary data to complete their own unique design tasks. However, this type of product is still far from the development goals of intelligent, automated, and collaborative CAE and CAD.

[0007] In order to solve the above problems, the present invention proposes natural language computer-aided engineering (NLCAE) and natural language computer-aided design (NLCAD) methods. By using artificial intelligence large language models and agents that have learned professional knowledge in the fields of CAE and CAD, CAE software and CAD software are driven to work together to achieve intelligent, automated and collaborative simulation analysis and design optimization. Summary of the invention

[0008] Purpose of the invention: Users use natural language, through artificial intelligence large language models and intelligent agents, to drive CAE software and CAD software to complete simulation analysis and design optimization tasks, realize the intelligence and automation of CAE and CAD, and enable product designers to complete complex simulation analysis and design optimization tasks through dialogue interaction with an intelligent agent, greatly reducing the threshold for design and innovation using CAE and CAD technologies.

[0009] Technical solution: To achieve simulation analysis and design optimization by driving CAE software and CAD software through natural language human-computer interaction, we must first have an artificial intelligence language model that can understand and generate natural language and program code, and let the language model further learn professional knowledge in the fields of CAE and CAD. The large language model can meet the requirements and can be fine-tuned with local data. People interact with the language model, CAE software and CAD software through an intelligent agent. The language model generates operating commands or executable program codes for CAE software and CAD software based on the simulation analysis and design requirements expressed by people in natural language. Then, the intelligent agent drives CAE software and CAD software to execute commands or programs to complete simulation analysis and design optimization tasks. Through this intelligent agent, collaboration across multiple CAE software and CAD software platforms can be more intelligent and automated. People do not need to switch back and forth between multiple software and only communicate with one intelligent agent. The specific steps are as follows: (1) Prepare a large language model and software and hardware environment: AI large language model training requires sufficient computing resources and training data. You can decide whether to train a large language model from scratch or choose a pre-trained open source large language model based on the actual situation. The large language model must support English. Currently, there are multiple open source large language models with excellent performance available for download. To use the open source model, you need to download the open source language model and the pre-trained model parameters to your local computer, prepare the necessary hardware and software environment for large language model training, and run and test the large language model. (2) Learn professional knowledge of CAE and CAD: Let the big language model learn professional knowledge in the fields of CAE and CAD so that it can correctly understand the simulation analysis and design requirements proposed by people and generate executable commands or program codes for CAE software and CAD software. The sources of CAE and CAD professional knowledge include but are not limited to help documents, manuals, tutorials, examples, relevant national standards and industry standards, papers, professional books, literature, articles on the Internet, etc. of various CAE software and CAD software, which need to convert text data into a format acceptable to the model before training. The basic steps include: installing the necessary deep learning framework and related libraries, loading the big language model and parameters described in the previous step, preparing CAE and CAD domain professional knowledge data, setting training parameters, splitting the training data set, training the model, evaluating and verifying the performance of the model, saving the trained model, and deploying the big language model as an API service; (3) Developing an agent based on a large language model: The large language model can understand and generate human language and program code. To perform more complex tasks, it is necessary to develop an agent based on the large language model. The agent calls the API service of the large language model. The user interacts with the large language model through the agent. The large language model generates software operation commands or program code, and then the agent drives the CAE software and CAD software to perform specific simulation analysis and design optimization steps. The agent based on the large language model needs to implement multiple functions to effectively handle complex tasks and interact with users and the environment. The basic functions include: managing the conversation context and handling complex interaction scenarios; calling the large language model API or executing code; knowledge retrieval and reasoning; having certain reasoning and problem-solving capabilities, providing decision-making suggestions or autonomous decision-making capabilities, and being able to complete specific tasks; multi-modal processing such as text, voice, and image; user interaction interface (conversation window, voice assistant, etc.); optimization and tuning for CAE and CAD tasks; real-time information acquisition capabilities, and being able to obtain the execution status and execution result information of CAE software and CAD software in real time; (4) Use the agent for simulation analysis and design optimization: deploy the aforementioned large language model and agent to the actual production environment. Users interact with the large language model through the agent. The large language model generates simulation analysis and design optimization solutions and specific implementation steps, and generates software operation commands or program codes. The agent then drives the CAE software and CAD software to perform specific simulation analysis and design optimization steps. Through the agent, users can use multiple professional CAE software and CAD software to collaboratively complete complex multi-physics field coupling simulation analysis and design optimization. During the dialogue between the user and the agent, the user can issue operation commands to the agent, or ask the agent about the problem-solving method flow, theoretical knowledge background, method application conditions, required input and output, etc. The user makes a choice and then issues an operation command, or the agent makes an autonomous decision to generate operation commands or program codes that can be executed by the CAE software and CAD software to drive the CAE software or CAD software to perform the corresponding operation. After completing the simulation analysis task, the user reads and analyzes the simulation results through the agent. If a design problem is found, the CAD software is driven to perform design changes. In the above simulation analysis and design optimization process, users, intelligent agents, CAE software and CAD software work together. Users can choose the easiest operation method and directly operate the software when it is easier to operate directly in the software interface. For repetitive or iterative simulation analysis and design optimization problems, the intelligent agent can complete them autonomously according to the pre-established solution steps and rules; (5) Continuously learn new knowledge and improve performance: During use, conduct reinforcement learning based on human feedback to continuously improve the performance of the large language model. Continue to collect new data, regularly fine-tune or retrain, so that the large language model can continue to learn new knowledge after deployment.

[0010] Beneficial effects: (1) CAE and CAD are indispensable tools for modern product design. The method proposed in the present invention greatly reduces the difficulty of using CAE, and can significantly improve design efficiency, improve quality, optimize performance, reduce costs, and enhance innovation; (2) Engineers can complete complex simulation analysis and design optimization tasks by simply communicating and interacting with intelligent agents in natural language; (3) Engineers no longer need to be proficient in the theoretical knowledge of various disciplines involved in CAE, complex software operations, and accumulate rich engineering experience to complete high-quality simulation analysis work; (4) Engineers can complete cross-disciplinary and cross-software platform multi-physical field coupling simulation analysis and design optimization through intelligent agents, and no longer need to be completed by multi-disciplinary engineers in collaboration, which greatly improves work efficiency; (5) The technical barriers between product design and simulation analysis are eliminated, so that design engineers can independently complete all simulation analysis and design optimization work. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a flow chart of the method of the present invention. Implementation

[0012] The technical solution of the present invention is further described below in conjunction with an embodiment. The natural language computer-aided engineering (NLCAE) and natural language computer-aided design (NLCAD) method described in this embodiment comprises the steps of: (1) Prepare a large language model and software and hardware environment: The various manuals and documentation of the current mainstream CAE software are mainly in English, and the large language model must support English. Select an open source large language model, taking the LLaMA large language model as an example, prepare the computer system environment, and take Ubuntu Linux as an example. The hardware needs to use GPU and install the drivers CUDA and cuDNN, install Python, and install PyTorch that supports CUDA. Download the LLaMA code library and weights, install dependent libraries, and run and test the model; (2) Learn professional knowledge of CAE and CAD: Obtain text and program code data from sources such as CAE and CAD software documents, manuals, tutorials, examples, relevant national standards and industry standards, papers, professional books, literature, articles on the Internet, etc. to ensure data quality and diversity. Clean the data, remove irrelevant content, standardize the format, remove duplicate data, and then format the data. Preprocess the data, convert the data into a format acceptable to the large language model, and split it into training sets, validation sets, and test sets. Use the prepared data to fine-tune the model. Fine-tuning and prompt engineering can be used in combination. Prompt engineering guides the model to generate professional content by designing domain-related prompt words. The two have their own advantages and can complement each other when used together. This combination can usually significantly improve model performance. Evaluate model performance on the test set and adjust model hyperparameters to optimize the model. Finally, save the fine-tuned model and deploy the model as an API service; (3) Developing an agent based on a large language model: The agent calls the API service of the large language model. The agent needs to implement multiple functions to effectively handle complex tasks and interact with users and the environment. The basic functions include: managing the conversation context and handling complex interaction scenarios; calling the large language model API or executing code; knowledge retrieval and reasoning; having certain reasoning and problem-solving capabilities, providing decision-making suggestions or autonomous decision-making capabilities, and being able to complete specific tasks; multi-modal processing such as text, voice, and image; user interaction interface (conversation window, voice assistant, etc.); optimization and tuning for CAE and CAD tasks; real-time information acquisition capabilities, and being able to obtain the execution status and execution result information of CAE software and CAD software in real time. The design of the agent mainly includes: designing a dialogue management module to manage the dialogue status and process; designing a multi-modal processing module to process various types of data; and designing an environment for the interaction of the agent. Currently, the commonly used development frameworks are LangChain, Transformers, or using SDK. The deployment method can be cloud deployment or local deployment. The main function achieved is that the user communicates with the large language model through the intelligent agent, and the large language model generates software operation commands or program codes, and then the intelligent agent drives the CAE software and CAD software to execute operations or programs; (4) Use intelligent agents for simulation analysis and design optimization: deploy the aforementioned large language model and intelligent agents to the actual application production environment. Users interact with the large language model through intelligent agents. The intelligent agents call the large language model to understand the task requirements proposed by the users, and provide feedback to the users on the operations required to complete the tasks, the physical models used, the applicable conditions of the physical models, and the model parameters that need to be input. During the interactive dialogue, the simulation analysis plan and the design optimization task step details are improved. The large language model generates CAE and CAD software operation commands or executable program codes, and the intelligent agents drive CAE software and CAD software to perform specific simulation analysis and design optimization steps. Users use intelligent agents to collaborate with multiple interdisciplinary professional CAE software to complete complex multi-physics field coupling simulation analysis. For example, fluid mechanics analysis CAE software and structural mechanics analysis CAE software collaborate to complete fluid-solid coupling analysis. During the conversation between the user and the agent, the user can issue operation commands to the agent or ask questions to the agent. The large language model generates the problem-solving method flow, theoretical knowledge background, applicable conditions of the method, required input and output, etc. The user makes a choice and issues an operation command, or the agent makes an autonomous decision to generate operation commands or program codes that can be executed by CAE software and CAD software to drive CAE software or CAD software to perform corresponding operations. After completing the simulation analysis, the user reads and interprets the analysis results through the agent and draws an analysis conclusion. If a design problem is found, the CAD software is driven to perform design changes. In the above-mentioned simulation analysis and design optimization process, the user, the agent, CAE software and CAD software collaborate with each other. The user can choose the simplest operation method and directly operate the software when it is easier to operate directly in the software interface. For repetitive simulation analysis and design optimization problems that require cyclic iteration, the agent can complete them autonomously according to pre-established solution steps and rules; (5) Continuously learn new knowledge and improve performance: During use, conduct reinforcement learning based on human feedback to continuously improve the performance of the large language model. Continuously collect new knowledge for the large language model, regularly fine-tune or retrain, continuously improve the performance of the large language model, improve its ability to understand professional technical terms and requirements, and improve the accuracy of performing simulation analysis tasks.

[0013] The above-described embodiment is only a specific implementation example of the present invention, and is not intended to limit the scope of implementation of the present invention. All changes made based on the principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A natural language computer-aided engineering and computer-aided design method, characterized in that: Includes steps: (1) Prepare the software and hardware environment required for large artificial intelligence language models and model training; (2) Fine-tune the large language model using professional knowledge data in the CAE and CAD fields, save the fine-tuned large language model, and deploy it as an API service; (3) Develop an intelligent agent based on this large language model; (4) The user interacts with the intelligent agent in natural language, calls the API service of the large language model, generates operation commands or program codes that can be executed by CAE software and CAD software, and the intelligent agent drives the CAE software and CAD software to execute commands or programs to complete simulation analysis and design optimization tasks; (5) During use, the performance of the large language model in CAE and CAD is continuously improved through reinforcement learning and regular fine-tuning with new data.

2. A natural language computer-aided engineering and computer-aided design method according to claim 1, characterized in that: In step (1), there are two ways to prepare a large language model: building a large language model from scratch and training it, or using a pre-trained open source large language model.

3. A natural language computer-aided engineering and computer-aided design method according to claim 1, characterized in that: In step (2), the fine-tuned large language model has the ability to understand CAE and CAD professional languages ​​and program codes and to generate operational commands and program codes executable by CAE software and CAD software.

4. A natural language computer-aided engineering and computer-aided design method according to claim 1, characterized in that: In step (3), the intelligent agent has the ability to obtain the execution status and execution result information and data of the CAE software and the CAD software in real time.

5. A natural language computer-aided engineering and computer-aided design method according to claim 1, characterized in that: In step (3), the intelligent agent has the ability to perform logical reasoning and solve specific problems based on the context, and the ability to provide decision suggestions or make autonomous decisions based on CAE and CAD data and context.

6. A natural language computer-aided engineering and computer-aided design method according to claim 1, characterized in that: In step (4), the user can interact with the intelligent agent to coordinate multiple professional CAE software and CAD software to perform multi-physics field coupling simulation analysis and design optimization.

7. A natural language computer-aided engineering and computer-aided design method according to claim 1, characterized in that: In step (4), for problems that are repetitive and require iterative simulation analysis and design optimization, the intelligent agent has the ability to autonomously complete the simulation analysis and design optimization tasks.

Images