Experiment. Learn. Build. Then scale — anywhere.
PINNeAPPle is an open-source Physics AI research and experimentation platform designed to take you from your first physics-informed neural network all the way to robust, production-ready solutions — independent of any specific framework, vendor, or ecosystem.
 |
 |
| Clamped Plate — deflection, Von Mises stress & bending moment | 2D Heat Equation — Exact vs PINN across time steps |
 |
 |
| Lamb-Oseen Vortex Pair — vorticity evolution | Allen-Cahn Phase Separation — interface dynamics |
Modern Physics AI ecosystems are powerful — but they assume you already understand:
- How to formulate physical problems correctly
- Which architectures to use (PINNs, operators, surrogates…)
- How to validate physics consistency
- How to benchmark and trust your results
PINNeAPPle is where you build that foundation.
Think of it as your experimentation layer:
Your physics problem
↓
[ PINNeAPPle ] ← experiment freely here
Understand the physics
Try architectures
Compare approaches
Validate results
Build intuition
↓
[ Your Target Stack ]
(custom infra, HPC, cloud, internal platform, etc.)
Scale, deploy, integrate
"I understand the physics. I want to see what AI can do with it."
from pinneaple_environment import BurgersPreset
problem = BurgersPreset(nu=0.01)
model = problem.build_model()
trainer = problem.build_trainer(n_epochs=3000)
result = trainer.fit(model)
result.plot()"I want to test ideas and compare approaches."
from pinneaple_arena import ArenaRunner
runner = ArenaRunner.from_yaml("configs/arena/burgers_benchmark.yaml")
results = runner.run_all()
results.leaderboard()
Potential Flow Past Circular Cylinder — exact solution vs PINN vs pointwise error
"I want to turn this into a real system."
from pinneaple_train.distributed import DDPPINNTrainer
from pinneaple_export.onnx_exporter import ONNXExporter
Multi-model forecast comparison across test windows — Naive, FFT-only, LSTM, FFT+LSTM
PINNeAPPle is stack-agnostic.
It helps you:
- Validate physics and modeling
- Benchmark architectures
- Export models
- Integrate with:
- HPC clusters
- Cloud ML pipelines
- Simulation frameworks
- Internal platforms
- Digital twins
| Module | Description |
|---|---|
| 📦 Unified Physical Data (UPD) | State, geometry, equations, metadata |
| 🌍 Data Pipeline | Zarr datasets, active learning |
| 📐 Geometry & Mesh | CAD, meshing, simulation interoperability |
| 🧠 Model Zoo | PINNs, Neural Operators, GNNs, Transformers, ROM, Classical |
| 🧮 Physics Loss Factory | Symbolic PDE → residuals |
| ⚙️ Solvers | FEM, FDM, FVM, Spectral, SPH, LBM |
| 🏗️ Training | Distributed, AMP, deterministic |
| 📊 Uncertainty & Validation | Dropout, ensembles, conservation checks |
| 🔁 Transfer & Meta-Learning | Fine-tuning, cross-PDE adaptation |
| 🛰️ Digital Twins | EnKF, real-time estimation |
| 🚢 Deployment | FastAPI, ONNX, TorchScript |
| 🤖 Problem Design | NLP → PDE |
| 🏆 Benchmarking | YAML, leaderboards |
If you can't validate it, you shouldn't deploy it.
Physics AI is about:
- ✅ Correct formulations
- ✅ Reliable validation
- ✅ Understanding failure modes
- ✅ Making informed decisions
| PINNeAPPle | |
|---|---|
| Vendor lock-in | ❌ Not tied to any vendor |
| Just a PINN library | ❌ Much more than that |
| Just experimentation | ❌ Bridges to production |
| ✅ What it is | A controlled environment to design, test, and validate Physics AI systems |
If this project makes sense to you, give it a star ⭐
It helps:
- Grow the ecosystem
- Attract contributors
- Build a real standard
Built for researchers and engineers who take physics seriously.