| name | moldgen-3d-mesh |
| description | Work with 3D mesh processing in MoldGen. Use when implementing geometry algorithms, mesh operations, mold building, orientation analysis, parting surfaces, or GPU kernels. |
MoldGen 3D Mesh Processing
Core Libraries
- trimesh: Primary mesh representation.
trimesh.Trimesh for vertices/faces.
- manifold3d: Fast boolean operations (union, difference, intersection).
- scipy.spatial: KD-trees, convex hull, spatial transforms.
- scikit-image: Marching cubes for SDF→mesh conversion.
- numpy: All geometry math operates on numpy arrays.
- numba: CUDA JIT for GPU-accelerated voxelization and ray casting.
Key Modules
moldgen/core/
├── mesh_io.py # Load/save (STL, OBJ, PLY, FBX, STEP, 3MF)
├── mesh_repair.py # Hole filling, normal fixing, degenerate removal
├── mesh_editor.py # Boolean ops, transform, simplify, subdivide
├── mesh_data.py # MeshData dataclass, quality metrics
├── orientation.py # Fibonacci sampling, undercut scoring, GPU ray casting
├── parting.py # Parting line/surface generation
├── mold_builder.py # Shell construction, alignment pins, bolt holes
├── gating.py # Pour hole, vent placement
├── insert_generator.py # Support plate: cross-section, conformal, anchor patterns
├── flow_sim.py # L1 heuristic + L2 Darcy flow simulation
├── fea.py # Finite element analysis
├── material.py # Material database (silicone, PU, epoxy)
└── optimizer.py # Gate position and parameter optimization
moldgen/gpu/
├── device.py # GPU detection (CUDA, Numba, CuPy)
├── sdf.py # SDF voxelization (numba CUDA)
├── ray_cast.py # GPU ray casting for orientation
├── flow_kernel.py # GPU Darcy flow solver
└── fallback.py # CPU fallback implementations
Patterns
- Functions take
trimesh.Trimesh or numpy arrays as input
- Return structured results (dataclasses or dicts)
- GPU functions check
GPUDevice().info.available and fall back to CPU
- Voxel grids: numpy 3D arrays, resolution typically 64-256
- Coordinate system: Z-up, millimeters
- Mesh quality: always check watertight before boolean operations
