Aerodynamics meets machine learning. This repository gathers my self-driven work toward becoming a CFD × Physics-Informed Machine Learning engineer: high-fidelity CFD simulations, and neural surrogate models that learn from them to predict aerodynamics in real time.
INSA Toulouse (Mechanical Engineering) · Erasmus @ Linköping University — Fall 2026.
- 🏎️ Formula Student wing aero explorer — set a wing element's angle and the car speed → downforce, drag and efficiency instantly, plus the optimal angle. A neural surrogate (NACA 4412 inverted) + a finite-wing induced-drag model, fully client-side. (A fast 2-D estimate for early design trade-offs — not a 3-D CFD.)
- 🛩️ NACA 0012 surrogate — move the angle (α) and Reynolds (Re) sliders and watch lift/drag/L-D update; the airfoil tilts with α. A PyTorch model exported to JavaScript.
cfd-projects/
├── piml/ # CFD × ML roadmap (the core of this repo)
│ ├── phase0_post_processor/ ✅ CFD post-processor + Verification & Validation
│ ├── phase1_surrogate/ ✅ neural surrogate (α, Re) → (Cl, Cd)
│ ├── phase2_pinns/ ✅ physics-informed neural networks (ODE → Navier–Stokes → airfoil)
│ ├── phase3_optimization/ ◻ shape optimization + dashboard (next)
│ ├── fs_wing_surrogate/ 🏎️ Formula Student wing aero explorer (live tool)
│ ├── applied_pinns/ 🧩 standalone PINN showcases (e.g. wing-spar deflection)
│ └── references/ 📚 PINN / Scientific-ML knowledge base (bibliography + playbook)
├── web/ # the live demo site (Vite + React)
├── naca0012-airfoil-CFD/ # standalone CFD study — NACA 0012 airfoil
├── front-wing-CFD/ # standalone CFD study — Formula Student front wing
└── docs/ # method memos + a full zero→expert PyTorch guide
| Phase | Focus | Status |
|---|---|---|
| 0 | CFD post-processor + Verification & Validation | ✅ done |
| 1 | Surrogate model (α, Re) → (Cl, Cd) — PyTorch, runs in the browser | ✅ done |
| 2 | Physics-Informed NNs — ODE → PDEs → inverse → 2-D Navier–Stokes → airfoil → parametric PINN (validated vs a panel method) | ✅ done |
| 3 | Shape optimization + dashboard | ◻ planned |
→ Start here: piml/
naca0012-airfoil-CFD/— symmetric NACA 0012 airfoil, α = 15°, Re ≈ 4·10⁵.front-wing-CFD/— Formula Student dual-element front wing (NACA 4412 inverted), Cl/Cd ≈ 3.19.
| Domain | Stack |
|---|---|
| CAD | Fusion 360, CATIA V5 |
| CFD | ANSYS Fluent 2026 R1, ANSYS Meshing |
| ML | Python · PyTorch · NeuralFoil / AeroSandbox |
| Web | Vite · React · TypeScript · Tailwind CSS · GSAP |
docs/pytorch_guide.md— a self-contained zero → expert PyTorch guide.piml/references/— curated PINN / Scientific-ML knowledge base (annotated bibliography + a practical playbook).docs/fluent_reports_NACA0012.md·docs/fluent_domain_mesh_NACA0012.md— Fluent setup memos.
Patxi Sallaberry — INSA Toulouse · Erasmus @ Linköping University (LiU), Fall 2026 LinkedIn · GitHub