Artificial Intelligence and Large Language Models in CAD

Abstract

The process of Finite Element Analysis (FEA) of components necessitates a CAD model that has been appropriately tagged and prepared. This tagging involves identifying and selecting specific parts of the CAD geometry for the application of boundary conditions, forces, or mesh specifications. While manual tagging is feasible for simple geometries, it becomes extremely complex and time-consuming for the intricate and large-scale components used by GKN Aerospace. To address this challenge, GKN Aerospace developed an in-house automated tool to streamline certain aspects of the tagging process, thereby saving time and effort. However, the tool requires proficiency in YAML, a programming language unfamiliar to many analysis engineers, complicating its usage and highlighting the need for a more user-friendly interface. The aim of this project was to explore the capabilities of Artificial Intelligence and Large Language Models in assisting with tasks such as comprehending and translating natural language terminology of geometric features and using the capabilities of these generative models to create extraction queries in SQL formatted in YAML. It is hypothesized that the capabilities of generative models, text in this case, could be useful in making the geometry tagging process used by the automated tool much more streamlined and simple to use, as it only requires natural language input and not complex code that may be subject to formatting errors. While there are multiple ways to carry out the tagging process, this project focuses only on the process used by the "Autotag" automated tagging tool. A set of research questions were formulated to guide the project and ensure it was focused on the right outcomes. The project was conducted over a period of 4 months. during the initial phases of the project, interviews were conducted to better understand the challenges and issues currently being faced by the engineers and GKN. Literature reviews were also conducted to gather information about the advantages and disadvantages about LLMs and AI and ways in which these technologies could be used to solve the issues that were identified. An application was then developed that addressed these problems, and was integrated into the "Autotag" tool in Sim-Center. The development, carried out with knowledge from literature reviews and GKN engineer interviews utilizes Large Language Models (LLMs) to simplify feature selection and identification. It allows use of natural language to ease communication between the user and the application. While limitations exist due to LLMs and complex CAD models, the project showcased a possible way in which LLMs can be used to augment the tagging process where manual selection may not be possible.