Conferences, 17th International Conference on Computational Methods

Font Size: 
Research on a CAD system for large-scale simulations using generative AI
YUZE LI, Ryuji Shioya

Last modified: 2026-07-30

Abstract


Large-scale numerical simulations are widely used to reproduce and predict complex physical phenomena in aerospace engineering, climate science, disaster mitigation, and other fields. However, preparing CAD models and analysis-ready meshes requires considerable time and human effort. This study develops an AI-assisted CAD system to reduce the workload involved in geometric modeling and mesh preparation for large-scale simulations. In recent years, generative AI has been adopted in many fields for reasons such as increased efficiency and cost reduction, and CAD software equipped with AI capabilities has also been developed. While AI-powered CAD software is expected to support human work and reduce working time, However, current AI-powered CAD systems still face challenges related to generation accuracy and the creation of complex geometries. This study develops an AI-assisted CAD system intended to reduce the workload involved in preparing geometric models and analysis-ready meshes. In the proposed system, a user describes the target geometry and dimensional constraints using text or images. Generative AI then produces Python code for FreeCAD or, for printed circuit board applications, a KiCad-compatible design file. The generated model is inspected and, when necessary, corrected in FreeCAD or KiCad. Mechanical models are exported in STL format, converted into MSH mesh files using Gmsh and Python, and analyzed with the ADVENTURE system, a large-scale CAE system developed at the University of Tokyo. Simulation results are converted to VTK format and examined in ParaView. In a preliminary experiment, three AI models each generated 48 CAD models. Claude Sonnet 5 achieved the highest overall accuracy. For simple geometries, the AI-assisted workflow reduced modeling time from several tens of minutes to less than 10 minutes. These results demonstrate the feasibility of connecting generative AI, open-source CAD tools, mesh generation, and large-scale CAE in a unified workflow, while also identifying the reliability improvements required for practical use.


Keywords


Generative AI, CAD, Simulation, System Construction

Conference registration is required in order to view papers.