Yugyeong Jang    Information    Contact
     
PUBLICATION

BIGcad: Assisting 3D CAD Modeling with Workflow Graph-
Driven Bayesian Command Inference





      Kyung Hoon Hyun, Yugyeong Jang,
      Archives of Design Research

    paper  

    Advancements in 3D Generative AI have greatly improved design capabilities for mesh-based 3D objects and environment modeling. Yet, modifying and editing these mesh-based models in detail is challenging. Designers continue to rely on command-based CAD tools like Rhino for precise and complex 3D modeling, crucial for fabrication and manufacturing that demands high accuracy and significant mental effort. To address this, we propose a framework combining a Bayesian inference model and workflow graph to predict user intentions, suggesting tailored commands and sequences to improve precision and efficiency. The framework was developed as a Rhino plugin to support users’ 3D modeling processes. User studies show our system reduces errors, streamlines workflows, and offers effective command recommendations.



    Generating Command Modeling and Design Graphs with Data Augmentation for Enhanced 3D Modeling Support






        Yugyeong Jang, Kyung Hoon Hyun
        Advanced Engineering Informatics

      paper  

      This study proposes a system that automatically generates 3D modeling sequences for various 3D shapes. Existing 3D modeling systems impose a high cognitive load on users, making it particularly difficult for beginners to approach. To address this issue, we developed a system that applies a method for inferring and extracting modeling sequences from 3D shapes to generate Command Modeling and Design Graphs without the need for additional modeling data collection. For this purpose, we reconstructed geometric elements and their structural relationships using a domain-specific language, efficiently modeling shape repetitions and symmetries. The proposed system infers modeling sequences from completed 3D models and converts them into workflow graphs, providing richer and more detailed sequence data compared to existing datasets. As a result, users are expected to significantly improve design efficiency through intuitive modeling processes and command support.