ecopus/mesh-refinement-agent

1---2title: Mesh Refinement Agent3emoji: 🏢4colorFrom: indigo5colorTo: gray6sdk: docker7pinned: false8short_description: This is an AI agent which uses optimizes mesh generation9---10 11Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference12 13# Mesh Refinement Agent (HF Space)14 15This Hugging Face Space runs a full machine-learning–guided mesh refinement workflow16for STEP CAD geometry.17 18The system performs:19 20- STEP loading and feature extraction  21- Detection of planes, fillets, circular holes, non-circular pockets  22- Part-level feature aggregation  23- Four ML model predictions:24  - Convergence classifier  25  - Global mesh size regressor  26  - Local refinement flag classifier  27  - Local mesh size regressor  28- Gmsh meshing with background field refinement  29- PNG + interactive Plotly visualization  30- `.msh` mesh export31 32---33 34## How to Use35 361. Upload a `.step` / `.stp` file  372. Select the load type (bend or tension)  383. Pick visualization mode:39   - Wireframe40   - Highlight (local refinement)41   - Heatmap (element size inverse)  424. Press **Run**43 44You will receive:45 46- A natural-language summary of predicted convergence & refinement  47- Global mesh PNG  48- Refined mesh PNG  49- Interactive 3D mesh viewer  50- Downloadable `mesh_refined.msh`51 52---53 54## Project Files55 56Key files included in this Space:57 58Models are pre-trained and do not require Ansys or dataset files.59 60---61 62## Notes63 64- Computation is CPU-only (HF Spaces restriction)65- Very large STEP files may take 10–40 seconds66- All geometry processing uses `cadquery-ocp` OpenCascade backend67- Gmsh 4.11.1 CLI is used for meshing68 69---70 71## License72 73MIT License74