Skip to content

MODEL_CARD.md

Latest commit

 

History

History
202 lines (149 loc) · 7.96 KB

MODEL_CARD.md

File metadata and controls

202 lines (149 loc) · 7.96 KB

Model Card — PEM-EC Designer

Scientific description of the simulation model, its scope, assumptions, and validity. Inspired by the Model Cards framework (Mitchell et al., 2019) adapted for physics-based simulation rather than ML.

Version: 0.3.0 Date: 2026-04-20 Authors: Abed Qadi (with AI-assisted implementation)

1. Intended use

Primary use cases

  • Early-stage PEM-EC cell sizing — pick membrane, catalysts, area, operating point
  • Stack-level performance estimation — voltage, power, hydrogen output for given N cells
  • Thermal management sizing — waste heat, cooling-loop requirement
  • Sensitivity analysis — effect of temperature, pressure, material choice on performance
  • Engineering education — visualize where the voltage losses go

Not intended for

  • Final design validation (use ANSYS Fluent, COMSOL, STAR-CCM+, AVL Fire M)
  • Degraded-cell performance prediction (no aging model)
  • Transient behavior (start-up, load-following, shutdown)
  • Local flow-field optimization
  • Safety certification
  • Commercial product warranty calculations

2. Model scope & approach

Physical dimensionality: 0D (lumped, zero-dimensional)

All quantities are treated as uniform across the cell. No spatial variation. Stack extension adds aggregation but no spatial coupling between cells.

Steady-state

No time dependence. Operating point is assumed instantaneously achieved and stable.

Governing equations

Domain Equation Reference
Thermodynamics Nernst (T + p corrected) Larminie & Dicks (2003), Eq. 2.22
Kinetics Full Butler-Volmer (Newton-inverted, no Tafel approx.) Bard & Faulkner (2001), Eq. 3.3.11; ADR 005
j₀ temperature dep. Arrhenius: j₀(T) = j₀_ref · exp(−E_a/R·(1/T − 1/T_ref)) Carmo et al. (2013), Eq. 9; ADR 004
Membrane conductivity Springer (1991): σ(λ, T) = (0.005139·λ − 0.00326)·exp(1268·(1/303 − 1/T)) Springer et al. (1991), Eq. 23; ADR 003
Ohmic transport Ohm's law, area-specific Barbir (2012), Ch. 2
Mass transport Concentration overpotential Larminie & Dicks (2003), Eq. 3.36
Energy balance Thermoneutral-voltage formulation Tijani et al. (2018), IJHE 43
Stack aggregation Electrical series Barbir (2012), Ch. 6

3. Assumptions (explicit)

Always assumed

  • Isothermal cell (no internal temperature gradient)
  • Isobaric cell (anode and cathode at same pressure)
  • Liquid water supply (water activity ≈ 1 on the anode side)
  • Constant Faraday efficiency (default 98 %)
  • Single rate-determining electron-transfer step per electrode (α_forward + α_backward = 1 in the Butler-Volmer form)
  • No gas crossover effects on voltage
  • Ideal gas behavior for Nernst pressure term
  • Linear stack aggregation (all cells identical, no inhomogeneity)
  • No bubble effects on electrolyte conductivity

Assumptions that users can override

  • Membrane thickness, λ_max, E_w → via material preset or explicit value
  • Membrane hydration ratio λ/λ_max → UI slider (0.30–1.00), feeds Springer σ(λ, T)
  • Catalyst j₀_ref, α, E_a (activation energy) → via material preset; j₀(T) auto-corrected
  • GDL resistance → via material preset
  • Operating T, p, j → via sidebar sliders
  • Coolant cp, ΔT → via sidebar

4. Validation status

Green (validated)

Check Method Tolerance Status
Tafel slope at high j vs. analytical b = 2.303·R·T/(α·F) — BV→Tafel limit < 0.5 % ✅ passing
Butler-Volmer roundtrip η = solve(j) → j = BV(η) recovers j < 1e-8 ✅ passing
BV → Tafel at high η j_BV − j_Tafel / j_Tafel at η = 0.5 V
Springer σ at 303 K, λ = 22 vs. literature 10.98 S/m rel_tol 1e-9 ✅ passing
Springer T-ratio σ(353)/σ(303) vs. Arrhenius-like 1.6–2.0 bracket ✅ passing
Arrhenius IrO₂ drop at RT j₀(298)/j₀(353) with E_a=52 kJ/mol 0.03–0.08 ✅ passing
E_rev at standard conditions vs. 1.229 V literature abs_tol 1e-9 ✅ passing
dE_rev/dT vs. -0.846 mV/K literature rel_tol 1e-6 ✅ passing
U_cell at 80 °C, 10 bar, 1 A/cm² vs. Carmo 2013 Fig. 6 1.6–2.2 V plausible ✅ passing
Stack linearity in N U_stack = N·U_cell rel_tol 1e-12 ✅ passing
Heat-balance conservation ṁ·cp·ΔT = Q rel_tol 1e-12 ✅ passing
H₂ scaling with area linear rel_tol 1e-9 ✅ passing

Amber (plausible but not experimentally validated)

  • Absolute polarization curve — values fall within literature range but have not been compared point-by-point to a specific experiment
  • Thermal efficiency — formula correct, magnitudes plausible, no experimental comparison yet
  • Cooling-flow predictions — consistent with energy balance, no field-data comparison

Red (not validated)

  • Concentration overpotential magnitude — uses a generic limiting-current form, not calibrated to a specific catalyst/GDL combination
  • Multi-phase effects — not modeled at all
  • Long-term performance — no aging mechanism

5. Known limitations & domain of validity

Operating range where the model is trustworthy

Parameter Valid range Confidence outside
Temperature 30–90 °C Membrane data thin below 25 °C; Arrhenius extrapolation OK within [273, 423] K
Pressure 1–30 bar Above 30 bar: ideal gas assumption deteriorates
Current density 0.01–2.5 A/cm² Full BV handles low-j regime; above 2.5: needs better j_lim
Membrane hydration λ > 5 (UI slider ≥ 0.30) Below λ ≈ 0.63: Springer yields non-physical σ, guard raises ValueError

Known systematic errors

  • E_a literature scatter ±15 % for some catalysts — translates to ±5 % j₀ error when |T − T_ref| > 30 K
  • Underpredicts ohmic losses if contact resistance is poorly specified
  • No correction for cell-to-cell variability in stack operation
  • No correction for pressure-drop across the flow field
  • Nafion-specific σ correlation applied to non-Nafion PFSA membranes (Aquivion, Fumapem) — accurate to ~10 % but not material-specific

6. Reproducibility

  • Deterministic — no random components
  • Version-pinned dependencies — see requirements.txt
  • Physical constants — CODATA 2018 (src/constants.py)
  • All equations sourced — @ref: annotations in every public function
  • Tests document expected behavior — 98 tests, all green

Given the same inputs and the same commit hash, the model produces byte-identical outputs.

7. Comparison to commercial tools

Capability This tool (v0.3) Full CFD (ANSYS/COMSOL/STAR-CCM+/AVL)
Cell-level U(j) accuracy ±5–10 % (unvalidated against exp.) ±1–3 %
Stack aggregation Linear, exact Can handle non-uniformity
Thermal (0D lumped) ±10 % on waste heat ±2 % with FEM
Runtime per point < 100 ms hours to days
Spatial fields (j(x,y), T(x,y)) ❌ none ✅ full
Transient ❌ ✅
Multi-phase ❌ ✅
3D geometry ❌ ✅
Cost / license $0, MIT €15k–70k/year + HPC

8. How to cite

@software{qadi_pem_ec_designer_2026,
  author  = {Qadi, Abed},
  title   = {{PEM-EC Designer: Physics-Based Simulation of PEM Water Electrolysis}},
  version = {0.3.0},
  year    = {2026},
  url     = {https://github.com/Tools00/pem-ec-0d}
}

9. Ethics & responsible use

  • This tool is for engineering education and early-stage design, not for safety-critical decisions
  • Do not rely on its outputs for commissioning real electrolyser systems without independent expert review
  • Do not use numerical results as-is in regulatory filings — commercial CFD and/or experimental data is required

If in doubt: ask a senior electrochemist.

10. Contact & feedback

Issues or questions → GitHub issue tracker (see README.md).