SimulationMusique

GPU-accelerated fluid dynamics simulation of a recorder flute using the Lattice Boltzmann Method. Generates acoustic signals by simulating air flow through a flute geometry defined in a PNG image.

Uses OpenCL for GPU computation, outputs VTK files for ParaView visualization, and records pressure data as audio.

Setup

Needs Python 3.8+ and an OpenCL GPU. Install with:

python3 -m venv env
source env/bin/activate
pip install -r requirements.txt

Running

python3 src/vonkarman_cl.py

Outputs VTK files to images/ and WAV to sounds/output.wav. Check the console for progress (prints every 400 steps).

Tweaking parameters

Edit src/vonkarman_cl.py:

• NU - viscosity (default 0.01). Higher = more stable but smoother
• N_iter - number of iterations (default 10,000,000)
• get_velocity() - inlet velocity. Currently ramps from 0 to 0.005

Geometry is defined in assets/simu_r2.png: red pixels are inlet, blue are outlet, black is walls, green is where audio gets sampled.

When things go wrong

If you get NaN values, the simulation diverged. Try:

• Lower velocity: Change the 0.05 in get_velocity() to 0.01
• Increase viscosity: NU = 0.05 instead of 0.01
• Narrow passages in the geometry can make things unstable

Looking at results

VTK files go to images/. Open them in ParaView.

Audio output goes to sounds/output.wav.