Computing Midcurve of a Thin Polygon for Mechanical Engineering
summary
Takes sheet‑metal 3D mid‑surface extraction and reduces it to computing the midcurve of a thin polygon (extruded sketch profile → 2D midcurve) and frames the input as graphs (vertices/edges) but solves it by converting data to images and JSON. Implements encoder–decoder / pix2pix style image-to-image translation and LLM fine‑tuning on serialized JSON serialization pairs, trains on augmented datasets, produces usable but noisy midcurves, and recalls mid‑surface meshing cuts solver time roughly to half.
AI-written summary · about
chapters (12)
- 0:00Introduction to talk
- 1:09Sheet metal in mechanical design
- 3:22CAD to CAE simplification
- 6:36Mid surface for thin parts
- 8:44Challenges in mid surface computation
- 10:53Reducing 3D to 2D mid curve
- 13:47Deep learning for mid curve
- 16:01Graph representation of shapes
- 18:32Graph to graph neural networks
- 20:24Image-based solution approach
- 21:27Text-based shape representation
- 23:56Summary and future work
related talks
18mFold, Paper, Scissors...
20mLightning Talk: Beyond Polygons: The Future of 3D Graphics with Implicit Models
01h 34mBootcamp on geometry and graph learning
42mDeep Learning approaches meet 3D data
11mInterview
35mAdjusting 2D prediction models to be usable for 3D objects
49mGraph Based Processing of Big Images
47m3D Digital Avatars with Machine Learning