-green.svg){kind=icon}
This project is a laptop-runnable 2D wind tunnel designed to bridge the gap between textbook aerodynamics and real-time visualization. Unlike standard game physics, this engine runs a Lattice Boltzmann Method (LBM) solver on the GPU, allowing it to simulate 200,000 interactive particles at 60 FPS.
It allows users to generate NACA 4-digit airfoils on the fly, visualize pressure/velocity fields, and perform automated angle-of-attack sweeps to generate professional Lift/Drag polar graphs.
- Method: Lattice Boltzmann Method (LBM) using the D2Q9 configuration (2 Dimensions, 9 Discrete Velocities).
- Engine: Taichi Lang (Python) for compiling high-performance CUDA kernels directly to the GPU.
- Visualization: Pygame for the GUI and surface rendering.
- Math: NumPy for data aggregation and airfoil geometry generation.
Why Taichi? Solving fluid dynamics requires iterating over hundreds of thousands of cells every frame. Standard Python is too slow for this. Taichi allows us to write Python-like syntax that compiles down to highly optimized GPU machine code, giving us C++ level performance with Python's flexibility.
This simulation was optimized to run on consumer hardware (laptops with dedicated GPUs). Key engineering challenges included:
- Massive Parallelism: Rendering 200,000 particles individually using a custom GPU kernel rather than CPU loops.
- Zero-Allocation Rendering: To prevent "Garbage Collection Stutter," all memory buffers (NumPy arrays and Pygame surfaces) are pre-allocated at startup. Frames are drawn by injecting data into existing memory slots rather than creating new objects.
- Numerical Stability: High-velocity fluid simulations are prone to "exploding" (values hitting infinity). We implemented strict CFL (Courant–Friedrichs–Lewy) conditions, limiting the lattice speed to maintain stability while using an Exponential Moving Average (EMA) to filter out high-frequency acoustic noise.
- Stair-Step Smoothing: Because the simulation runs on a pixel grid, curved airfoils suffer from "voxelization artifacts" that trap fluid. We calibrated the viscosity and smoothing algorithms to mitigate these spikes in drag/lift data.
-
Automated Data Sweeps: Press 'D' to initiate a full autonomous sweep from -5° to +20° Angle of Attack. The system waits for convergence, records
$C_l$ and$C_d$ , and rotates the wing automatically. - Professional Polar Plots: Generates a real-time Lift vs. Drag polar graph. Click the graph to expand it into a detailed scientific plot with axes, ticks, and calculated Max L/D Ratio.
-
Multi-Modal Visualization:
- Speed: Heatmap of velocity magnitude.
- Curl: Visualizes vorticity and turbulence (red/blue).
- Pressure: Visualizes high (red) and low (blue) pressure zones (Bernoulli's Principle).
- Particles: 200k Lagrangian particles for flow visualization.
-
Live Geometry: Type any 4-digit code (e.g.,
2412,0010) to generate and test custom airfoils instantly.
| Key | Action |
|---|---|
| SPACE | Pause / Resume Simulation |
| R | Soft Reset (Clear Airflow) |
| C | Hard Reset (Clear Airflow & Obstacles) |
| A | Open Airfoil Menu (Type NACA Code) |
| D | Start Data Sweep |
| X | Cancel Active Sweep |
| 1 | View Mode: Curl (Vorticity) |
| 2 | View Mode: Speed (Velocity Magnitude) |
| 3 | View Mode: Particles (Flow Lines) |
| 4 | View Mode: Pressure (Density) |
| H | Toggle HUD |
| Click Graph | Expand/Collapse Scientific Plot |
This project relies on the open-source Python ecosystem:
- Taichi Lang: For the GPU physics backend.
- Pygame: For window management and rendering.
- NumPy: For vector math and array manipulation.
- Numba: For auxiliary CPU optimization.
Jace Hawkins Jan 2026