Design Dataset for Intelligent Data Exchange


During my internship at Autodesk, I proposed a novel way of design data exchange using machine learning(ML) within 3d modeling softwares that can predict the user intended topological structure of 3d geometry and as an initial step create a 3d dataset with labels of its compositional and relational information. This shed light on ML-based subsystems for Autodesk’s cloud platform Forge to facilitate design data exchange and allow interorperability of different design softwares. I devised a novel error-based approach to extract user annotations of dimensions and geometric constraints from Revit 3d models and output a graph that encapsulates the topology of the geometry and user intentions. While developing this work, I actively gained insight from teams in Generative Design, Human-Computer Interaction, and Artificial Intelligence. With this project, I propose a machine-learning based system infornation.


Autodesk’s Cloud Platform Forge and Design Data Exchange




Machine learning based solution



With this project, I propose a machine-learning based system that can predict the user intended topological structure of 3d geometry and as an initial step create a 3d dataset with labels of its compositional and relational infornation.




Design Intension to Graph Structure






Nodes are geometric elements (solids, reference planes) and edges are alignment constraints. Graph structures of various windows reveal potential representations of distinctive features which can be interpreted by a machine learning model.




I devise an error-based pipeline within Revit to extract the graph structure from 3d models.





©2022 Fall