First time here?

The fuel-cell twinin ten minutes

This site is a digital twin of hydrogen fuel cells: computer models of PEM and solid-oxide cells, kept honest against measurements and published data. Use it to size a fuel-cell system, compare the two chemistries, and see exactly how much of each number has been checked.

What a digital twin does

Measure, model, check, plan

  1. MeasureVoltage, current, temperature, gas flows and impedance from a test rig, or published data.
  2. Model the physicsEquations for the cell voltage, heat, water and gas, solved for the conditions measured.
  3. Check and quantify uncertaintyCompare predictions with data the model never saw, and keep every alternative that still fits.
  4. Plan the next testRank rig conditions by how much they would teach the model per minute of testing.
The twin loop: measure, model the physics, quantify uncertainty, plan the next test.

Where to start

A tour of the site

Fuel cells in two minutes

Two ways to turn hydrogen into electricity

PEM

  • About 80 °C, starts in under a minute, follows load.
  • Flat plates stacked in hundreds, like the Toyota Mirai’s 370-cell stack.
  • Low-grade heat: big radiators.
  • Main candidate for aircraft propulsion.

SOFC

  • 600–1,000 °C, slow to start, best at steady load.
  • Ceramic cells, often tubes; very efficient, tolerant of fuels.
  • High-grade heat that a turbine can use.
  • Candidate for APUs, ground power and hybrids.

How to read the results

Every number says how far to trust it

Status labels in the API and tables

MODEL_OUTPUT_NOT_VALIDATEDStraight from a twin model; no independent test confirms it.
SCREENING_ESTIMATEA sizing estimate for discussion: stack only, simplified losses. Not design data.
CALIBRATED_ONE_POINT_NOT_VALIDATEDAdjusted to match one published point (the Mirai-class PEM cell).
SCREENING_ASSUMPTION_NOT_VALIDATEDA stated screening assumption inside a published range (the reference and high-performance SOFC cells).
POSTERIOR_FROM_PUBLIC_DATA_NOT_VALIDATEDA Bayesian fit to public data, with uncertainty bands; still not a validation.

The model switches

Try it

Size a 2 MW regional aircraft in five minutes

  1. Open PEM vs SOFC for aircraft and leave the switch on Side by side.
  2. In the Fuel-cell explorer, choose Step 2 · 2 MW, a regional aircraft of 9–100 seats.
  3. Read the metrics table: PEM needs about 255 m² of cells and 96 kg/h of hydrogen; SOFC about 333 m² and 94 kg/h (default models).
  4. Turn the 3D view to see the stacks to scale beside a 1.8 m person; the grey boxes are the balance of plant.
  5. Switch PEM model to Lab cell and SOFC model to Lab rig cell: the same aircraft now needs far more cell area. That gap is why calibration matters.
  6. Scroll to Size your own case and type your own power, then check the Sources at the bottom.

Words you will see

Glossary

Digital twin
A computer model of a physical system that is kept in step with measurements from it, used to predict, explain and plan.
PEM fuel cell
Proton-exchange membrane cell. Runs near 80 °C, starts in under a minute, follows load; used in cars such as the Toyota Mirai.
SOFC
Solid-oxide fuel cell. Runs at 600–1,000 °C, more efficient and fuel-flexible, but slow to start; suits steady duty.
Cell, stack, BoP
A cell is one membrane and its electrodes; a stack is hundreds of cells in series; balance of plant (BoP) is everything else: compressor, cooling, humidifier, power electronics.
Current density (A/cm²)
Current per square centimetre of cell. Higher means more power from the same area, at a lower voltage.
Polarization curve
Cell voltage against current density. The basic fingerprint of a fuel cell.
Power density (W/cm²)
Voltage × current density: power per square centimetre of cell.
Efficiency (LHV)
Electrical energy out divided by the hydrogen’s lower heating value. Roughly cell voltage ÷ 1.25 V.
kW/kg
Power per kilogram, for the stack or the whole system. The key aviation metric.
BoP kW per stack kW
Power the balance of plant uses, mostly the air compressor, for each kW the stack makes.
Impedance (Nyquist plot)
How the cell responds to a small AC signal over frequency; separates resistance and electrode processes.
Calibration vs validation
Calibration adjusts a model to data. Validation tests it on data it never saw. This site keeps them apart.
Posterior, hold-out
A posterior is a model’s parameters with their uncertainty after seeing data. A hold-out check predicts data left out of the fit.

Questions

FAQ

Is the twin validated?

No. It is a development candidate. The evidence board shows exactly what has been checked; independent physical validation is pending.

Where do the numbers come from?

From the twin’s models, run when the site is built, and from public sources cited on each page. Each evidence card links to the file and its checksum.

Why can I switch the PEM and SOFC models?

The twin’s own cells are lab cells. The Mirai-class PEM shows a commercial stack cell; the SOFC reference and high-performance cells span the published 0.5–1 W/cm² range. The answer changes, so the page lets you see both.

Can I use it in my own work?

Yes, through the web API or by installing the code (pip install -e . then fctwin --help). Treat outputs as screening estimates.

What is not modelled?

Altitude, start-up energy, thermal cycling, mass from first principles, and certification. Balance-of-plant shapes in 3D are placeholders.