arriemeijer-creator

arriemeijer-creator / JAX-differentiable-CFD

Public

A fully differentiable, structure-preserving framework for incompressible flow simulation, control, and inverse design. Combines physics-based prediction with multi-scale neural correction, latent-space acceleration, and differentiable geometry optimization. Built with JAX and Equinox.

11
0
100% credibility
Found Mar 30, 2026 at 11 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

An interactive framework for simulating and visualizing incompressible fluid flows using traditional numerical methods, with modular components for easy experimentation and data generation.

How It Works

1
🔍 Discover the fluid flow simulator

You find this handy tool for watching air and water flows around objects like cylinders or cavities.

2
📥 Set it up easily

Download and prepare it on your computer in just a few minutes, no hassle.

3
🚀 Launch the colorful viewer

Open the window and immediately see live fluid motion with swirling patterns and graphs updating smoothly.

4
⚙️ Pick your flow adventure

Choose from vortex streets, spinning cavities, or swirling vortices, and tweak speed or detail level.

5
🌊 Watch magic unfold

See beautiful real-time animations of flows, forces, and energy with six live plots that tell the story.

6
💾 Capture your results

Record videos of the action or save data files to use in reports or experiments.

✅ Flow expert unlocked

You now have accurate simulations ready for design, research, or sharing your discoveries.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 11 to 11 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is JAX-differentiable-CFD?

This JAX-built framework runs fully differentiable incompressible flow simulations for classic cases like vortex shedding and lid-driven cavities. It generates high-fidelity ground truth data via validated physics solvers, while exposing hooks for neural corrections, latent-space acceleration, and geometry optimization. Developers get a real-time GUI to tweak Reynolds numbers, grids, and schemes, plus CSV exports ready for machine learning simulation workflows.

Why is it gaining traction?

Unlike rigid CFD tools, everything here is end-to-end differentiable, letting you backprop through physics for inverse design or neural operator training. The modular swaps between 9 advection schemes and 7 pressure solvers, backed by a 1,680-config test suite, make prototyping fast with GPU acceleration. Real-time visualization and adaptive timestepping handle complex flows without stability headaches.

Who should use this?

CFD engineers integrating ML for turbulence modeling or sensor placement. ML researchers training physics-informed nets on flow data. Optimization experts tackling differentiable geometry tweaks for control problems.

Verdict

Solid baseline for fully differentiable CFD experimentation, with working GUI and tests, but low maturity (11 stars, 1% credibility score) means neural features are WIP—prototype data generation now, contribute to unlock inverse design potential.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.