EAGER/Collaborative Research: An LLM-Powered Framework for G-Code Comprehension and Retrieval

这项早期概念探索性研究补助金（EAGER）支持研究，以实现专门为增材制造量身定制的新的人工智能智能数据库。这个名为G-Forge的数据库将包含大量的g代码文件、驱动3D打印机的指令，以及用于查询、推理和翻译这些文件的有效工具。这些功能将通过多模态大型语言模型（llm）的最新进展得以实现。与目前的技术相比，这种方法提供了许多优点，包括早期识别制造计划中的潜在错误和快速、逐步调试代码，从而减少错误和延迟。预计G-Forge一旦用于3D打印，就可以扩展到其他数控机床，以重新定义传统的工作流程，大大减少错误和成本。该项目的多学科组成部分将被整合到更广泛的教育工作中，为学生提供网络制造关键跨学科领域的坚实基础，并最终支持经济竞争力、国家安全和劳动力发展。G-Forge是一个用于增材制造的多模式法学硕士，将使用计算机辅助设计（CAD）模型和G-code进行培训。LLM可以被认为是一个信息编码器，它将来自各种模式输入数据的信息编译成嵌入。这种嵌入可以用于不同的下游任务，例如验证、调试和索引潜在的大量G-Code文件集。由于其多方面的实用性，预计制造商将广泛采用G-Forge，激励他们参与创建共享的G-code数据库。G-Forge将为类似于“谷歌制造”的更大生态系统奠定基础，使用户能够执行许多下游任务，如设计检索和推荐以及自动形状生成。G-Code验证和调试：G-Forge的核心组件将是一个llm驱动的工具，可以评估给定G-Code文件中指定的部件是否适用于特定机床；社区驱动的G-Code数据库：与G-Forge交互的用户将有助于创建一个大型的经过验证的G-Code库；作为服务的G-Code分析：随着社区使用量的增加，G-Forge数据库将随着时间的推移而增长，最终将成为专门为G-Code调整的多模态基础模型的有价值的训练数据来源。该研究的教育和推广计划包括：(1)在爱荷华州立大学（ISU）和纽约大学（NYU）开发制造证书课程中的人工智能，以及(2)开发网络物理系统辅修课程的现有课程模块和网络制造的新实验室模块。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This EArly-concept Grant for Exploratory Research (EAGER) award supports research to enable a new AI-powered smart database specifically tailored for additive manufacturing. The database, G-Forge, will consist of a vast set of G-code files, the instructions that drive 3D printers, along with efficient tools for querying, reasoning about, and translating those files. Those capabilities will be enabled by recent advances in multimodal large language models (LLMs). This approach affords many advantages over the current state of the art, including early identification of potential errors in manufacturing plans and fast, step-by-step debugging of code, thereby reducing errors and delays. It is expected that G-Forge, once demonstrated for 3D printing, can be extended to other numerically controlled machine tools to redefine traditional workflows, significantly reducing errors and costs. The project's multidisciplinary components will be integrated into a broader educational effort to offer students a solid foundation in the critical interdisciplinary area of cyber manufacturing and will ultimately support economic competitiveness, national security, and workforce development. G-Forge is a multimodal LLM for additive manufacturing that will be trained using computer-aided design (CAD) models and G-code. The LLM can be considered an information encoder that compiles the information from input data of various modalities into an embedding. That embedding can be used for different downstream tasks, such as verification, debugging, and indexing of a potentially vast set of G-Code files. Because of its multi-faceted utility, it is expected that manufacturers will widely adopt G-Forge, incentivizing them to participate in creating a shared G-code database. G-Forge will lay the foundations of a larger ecosystem akin to 'Google for Manufacturing,' enabling users to perform numerous downstream tasks such as design retrieval and recommendation and automated shape generation. This vision incorporates the following specific objectives: G-Code Verification and Debugging: the core component of G-Forge will be an LLM-powered tool that can assess whether the part specified in a given G-Code file is valid for a particular machine tool; A Community-Driven G-Code Database: Users interacting with G-Forge will aid in the creation of a large library of verified G-Code; G-Code Analysis as a Service: With increasing community usage, the G-Forge database will grow over time, eventually serving as a valuable source of training data for a multimodal foundation model specifically tuned for G-code. The education and outreach plans of the research include: (1) the development of AI in manufacturing certificate programs at Iowa State University (ISU) and New York University (NYU), and (2) developing modules for existing courses in a cyber-physical systems minor and a new lab module in cyber manufacturing.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.