ecopus/mesh-refinement-agent

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Mesh Refinement Agent (HF Space)

This Hugging Face Space runs a full machine-learning–guided mesh refinement workflow for STEP CAD geometry.

The system performs:

• —STEP loading and feature extraction
• —Detection of planes, fillets, circular holes, non-circular pockets
• —Part-level feature aggregation
• —Four ML model predictions:
• —Convergence classifier
• —Global mesh size regressor
• —Local refinement flag classifier
• —Local mesh size regressor
• —Gmsh meshing with background field refinement
• —PNG + interactive Plotly visualization
• —.msh mesh export

How to Use

1. 1.Upload a .step / .stp file
2. 2.Select the load type (bend or tension)
3. 3.Pick visualization mode:
4. 4.Wireframe
5. 5.Highlight (local refinement)
6. 6.Heatmap (element size inverse)
7. 7.Press Run

You will receive:

• —A natural-language summary of predicted convergence & refinement
• —Global mesh PNG
• —Refined mesh PNG
• —Interactive 3D mesh viewer
• —Downloadable mesh_refined.msh

Project Files

Key files included in this Space:

Models are pre-trained and do not require Ansys or dataset files.

Notes

• —Computation is CPU-only (HF Spaces restriction)
• —Very large STEP files may take 10–40 seconds
• —All geometry processing uses cadquery-ocp OpenCascade backend
• —Gmsh 4.11.1 CLI is used for meshing

License

MIT License