From your test data to engineering decisions.
A four-step pipeline: your data feeds a physics-based cell model, an AI surrogate trained on your data extends it, and the combined model points at the manufacturing parameters worth changing.
Step 01
Your data
Polarisation (IV) curves, electrochemical impedance spectra, durability and cycling tests, operating logs and manufacturing parameters — the data your lab and test benches already produce.
- IV curves
- EIS
- Durability tests
- Manufacturing parameters
Schematic illustration — no experimental data shown.
Step 02
Physics-based cell model
A physics-based model of your cell and stack captures electrochemistry, species transport and degradation mechanisms — grounded in the laws that govern your technology, not fitted curves alone.
- Electrochemistry
- Transport
- Degradation
Schematic illustration — no experimental data shown.
Step 03
AI surrogate model, trained on your data
A fast AI surrogate trained on your own data learns the residual behaviour the physics model cannot capture analytically — extending short tests into performance and lifetime predictions.
- Trained on your data
- Fast inference
- Uncertainty-aware
Schematic illustration — no experimental data shown.
Step 04
Decisions
The output is not a report — it points at the manufacturing parameters to change: which porosity, thickness or catalyst loading moves performance and lifetime, and by how much.
- Porosity
- Thickness
- Loading
- Performance & lifetime
Schematic illustration — no experimental data shown.
Move an input. Watch the decision change.
Adjust the example inputs and see them flow through the four steps. Illustrative toy model with made-up coefficients — not calibrated on any real cell or customer data.
01 · Example inputs
02 · Physics model
Kinetic, ohmic and mass-transport losses respond to your inputs.
03 · Surrogate prediction
Cell voltage
1.88 V
Lifetime index
89 / 120
04 · Decision
- porosity30.0 → 44.0
- thickness18.0 → 14.0
- loading1.0 → 1.9
Predicted: 1.76 V · lifetime index 120
Illustrative data only. For your own parameters, try the parameter advisor.
Fewer experiments, more confident conclusions.
Why physics + AI works even with limited industrial data: a physics-based model carries most of the structure — electrochemistry, transport, degradation — so the AI never has to learn cell behaviour from scratch. It only has to correct the gap between the ideal model and your real, imperfect hardware. That means a surrogate trained on a few well-instrumented tests can generalise far beyond the data it has seen, whereas a purely data-driven model would need orders of magnitude more experiments before its predictions could be trusted.
Start with the data you already have.