System architecture
Technology
Phone capture on the edge, CPU orchestration and CAD at the core, and burst GPU compute only when a heavy job earns it.
Live
Phone capture
Photos, video, LiDAR where available, the 1 mm metric mat, ArUco tags and hard caliper readings.
Live
CPU server
Ingest, part-shoot sorting, parametric CAD, provenance, orchestration and lightweight analysis.
Pilot
GPU burst pods
Rented NVIDIA CUDA pods for DALI, heavy vision/model workloads and experiments that should not live permanently on the VPS.
Geometry + simulation
The stack follows the engineering question
There is no single “AI model” pretending to be a mechanical engineering department. Geometry, numerical solvers and learning components each get a bounded job.
| Layer | Tools / methods | Job | Status |
|---|---|---|---|
| Capture | Phone photo/video/LiDAR · ArUco · 1 mm mat · calipers | Acquire geometry and scale with explicit provenance. | Live |
| Reconstruction | Structure-from-motion · silhouettes · revolve · planform+depth | Choose the simplest method that matches part topology. | Live |
| Implicit geometry | PicoGK · signed-distance fields · voxels | Lattices, channels, helicoids, log-spiral hubs, negative-space fluid geometry. | Live |
| Parametric CAD | CadQuery · build123d · FreeCAD | Editable source, STEP, dimensioned drawings. | Live |
| Structural / modal | CalculiX-oriented workflows · Goodman + Basquin | Static FEA, mode shapes, resonance and fatigue. | Live |
| CFD | OpenFOAM k-ω SST | Flow and pressure questions. | Live |
| Acoustics | 1-D transmission-loss / Helmholtz-oriented models | Exhaust sound decision support. | Live |
| GPU vision | NVIDIA DALI · PyTorch | GPU decode/augmentation and model inference/training. | Pilot |
| Resonance sweeps | Optuna · SALib · PyVista · meshio | Explore variables and rank mode sensitivity. | Pilot |
| Magnetic simulation | Harmonica · Verde · Boule · ppigrf · MagNav.jl | Forward-model anomalies and work with public magnetic-navigation data. | Roadmap |
| Biomagnetic research | MNE-Python | OPM/MEG-style signal processing research. | Roadmap |
R&D pod
GPU install reference
This mirrors the current research plan: CUDA 12-class rented pod; verify package versions on the pod before relying on them.
# GPU data pipeline + learning
pip install --extra-index-url https://pypi.nvidia.com nvidia-dali-cuda120
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121
# Parametric resonance mapping
pip install optuna SALib pyvista meshio
# Magnetic anomaly forward modelling
pip install harmonica verde boule ppigrf xarray pyproj
julia -e 'using Pkg; Pkg.add("MagNav")'
# Bio-magnetic research
pip install mne