Airfoil CFD Lab — WebGPU

Interactive 2D airfoil aerodynamics in the browser. Three solvers cover Mach 0–6, run entirely client-side (GitHub Pages friendly, no backend, no build step), and every result is checked against wind-tunnel data or exact theory in a built-in Validation tab.

Engines

Engine	Method	Regime	Physics
LBM	D2Q9 lattice-Boltzmann, BGK + Smagorinsky LES, momentum-exchange forces	M < 0.3	Viscous, unsteady — Reynolds number is the knob; vortex shedding, separation
Euler	Finite volume, MUSCL + minmod, HLLC, dimensional splitting, immersed boundary	M 0.3–6	Compressible, inviscid — shocks, expansion fans, transonic buffet, bow shocks
Theory	Hess-Smith panel method (+ Kármán-Tsien), Ackeret linearized theory, exact shock-expansion (diamond)	instant	No GPU needed; also powers reference overlays

Auto mode picks LBM below M 0.3 and Euler above. Both GPU engines rotate the inflow for angle of attack, so geometry changes never re-rasterize on an α change.

Features

• Airfoil database: NACA 4/5-digit generators (any code), Clark Y, RAE 2822, S1223, E387, S809 (UIUC coordinates), diamond wedge, flat plate, plus custom .dat paste/upload (Selig & Lednicer formats)
• Mach 0–6, Re 10³–10⁸, α −15°…+20°, three grid resolutions, live parameter changes
• Fields: velocity, vorticity, Cp, density, numerical schlieren, temperature, local Mach — viridis/inferno/diverging colormaps, tracer particles
• Plots: surface Cp (panel/Ackeret lines + solver samples), lift polar with wind-tunnel anchors, drag polar, force history; automatic α-sweep; CSV export
• Readouts: C_L, C_D, C_M(c/4), L/D with convergence detection
• Works without WebGPU too (Theory engine + CPU potential-flow rendering)

Validation (the point of this project)

Every converged run is compared automatically against the best available reference:

• NACA 0012 / 2412 / 4412 — Abbott & von Doenhoff (1959) wind-tunnel polars, Re 6×10⁶
• RAE 2822 Case 6 — AGARD AR-138 transonic benchmark (M 0.725, Cl 0.743)
• Diamond airfoil — exact oblique-shock + Prandtl-Meyer solution (supersonic)
• Ackeret / thin-airfoil / Prandtl-Glauert — linearized theory across the range
• Panel method — d'Alembert self-consistency, lift-slope cross-check

The CI-style test suite (node tests/run_all.mjs, 77 checks, no GPU required) validates the numerics themselves:

• HLLC + MUSCL flux math vs exact Riemann solution (Sod tube, L1 error < 0.002)
• LBM collision/streaming vs Taylor-Green vortex analytic decay (< 0.5% error)
• Panel method vs thin-airfoil theory and A&vD anchors (lift slope, α₀, C_m)
• Shock relations vs Anderson's tables (β, p₂/p₁, ν(M) exact to table precision)
• WGSL shader structural sanity

Honest limitations

Staircase immersed boundaries (~5–15% force error, improves with grid), no transition modeling, LBM under-resolves boundary layers above Re ≈ 10⁵ (C_D indicative; C_L fine), Euler is inviscid (use the supplied empirical friction C_D₀ for totals), calorically perfect gas (above M ≈ 4–5 treat as qualitative). The Validation tab states the trust level for every regime. Educational tool — not for certification.

Run locally

Any static server works (ES modules need http, not file://):

cd CFD-WebGPU
python -m http.server 8000     # or: npx serve
# open http://localhost:8000

WebGPU requires Chrome/Edge 113+, Safari 18+, or Firefox 141+.

Deploy to GitHub Pages

1. Create a GitHub repo and push this folder (main branch).
2. Repo Settings → Pages → Source: GitHub Actions — the included .github/workflows/deploy.yml publishes on every push. (Alternative: Source: Deploy from a branch → main / root. The .nojekyll file is already there.)
3. Open https://<user>.github.io/<repo>/.

Tests

node tests/run_all.mjs

Architecture

index.html, css/style.css      UI shell (no frameworks, no build step)
js/main.js                     orchestration, run loop, charts/validation wiring
js/airfoils.js, airfoil-data.js  NACA generators, UIUC coords, .dat parser, rasterizer
js/panel.js                    Hess-Smith panel method (CPU)
js/theory.js                   PG/KT corrections, Ackeret, oblique shock + PM, friction drag
js/lbm.js + shaders/lbm.wgsl   D2Q9 LBM engine (WebGPU compute)
js/euler.js + shaders/euler.wgsl  compressible Euler engine (WebGPU compute)
js/viz.js + shaders/render.wgsl   field rendering, colormaps, particles, readback
js/charts.js                   dependency-free canvas plots
js/validation.js               reference data + auto-comparison
js/cpuflow.js                  no-GPU potential-flow visualization
tests/                         headless validation suite (Node, no GPU)

Credits

Airfoil coordinates: UIUC Airfoil Coordinates Database (M. Selig). Wind-tunnel data: Abbott & von Doenhoff, Theory of Wing Sections (1959). RAE 2822: Cook, McDonald & Firmin, AGARD AR-138. Colormap fits: M. Zucker. Numerical methods: Toro (HLLC), Moran (panel), Anderson (compressible relations).

MIT License.