Case study · Energy

Reproducing a Simulation from an Academic Paper

High-temperature PEM fuel cell Electrochemical multiphysics with Argus & COMSOL

Tools

Argus · COMSOL Multiphysics 6.3

Outcome

Within 0.2% of published data · physics reasoned from scratch

Case study · Energy

What’s inside

A walk-through of how Argus set up and solved a fully coupled multiphysics model of a high-temperature proton exchange membrane (HT-PEM) fuel cell in COMSOL, and reproduced a published benchmark to within 0.2% at every point on the polarisation curve. See what Argus reasoned beyond output, where Argus is fast and exact, and where the engineer’s judgment stayed essential.

Summary

A fully coupled HT-PEM fuel-cell model, built and solved with Argus, validated to within 0.2% of published scientific data across its full performance curve.

Argus set up and solved a fully coupled 3D model of a high-temperature proton exchange membrane fuel cell in COMSOL. The agent applied electrochemistry, mass transport (four-species Maxwell–Stefan model), and fluid flow physics to the simulation.

≤0.2%

Deviation from scientific benchmark

Coupled domains, one solve

0.013%

Max charge mismatch across the voltage sweep

Problem

A high-temperature PEM fuel cell is one of the most tightly coupled problems in engineering: the electrochemistry, the gas transport and the flow all depend on one another, so none can be solved on its own. Such a model is slow to solve, and a solution can converge but still be physically wrong. The geometry comprises seven stacked domains: anode flow channel, anode gas diffusion layer (GDL), anode gas diffusion electrode (GDE), polymer membrane, cathode GDE, cathode GDL, and cathode flow channel.

Transport coupling limits the cathode

As current increases, the cathode’s oxygen demand rises, yet oxygen’s effective diffusivity through nitrogen and product water is far lower than hydrogen’s. The cathode therefore reaches mass-transport limitation before the anode, and the coupled model must resolve this balance at every point on the polarisation curve.

Convergence is not validity

A coupled model can converge without crashes, solver errors or warnings and still produce results that are physically incorrect. Establishing that a solution is physically valid, and not merely numerically converged, is the substantive task.

+ Benefit to engineer

Argus performed the simulation: seven domains, approximately thirty domain and boundary settings across three physics interfaces, and a five-step solver configuration, with every step logged for traceability. Argus also supplies the physics reasoning behind the results, so the engineer can concentrate on validating the converged result.

Methodology

The model is one repeating channel of the cell, seven stacked layers from anode to cathode, with a phosphoric-acid-doped PBI membrane operating at 180°C. The engineer’s first call was to build it in 3D rather than 2D, because the current varies sideways across the rib; that in-plane (X–Y) variation is one a 2D down-channel slice (Y–Z) cannot produce. Before seeing any paper, Argus proposed the physics from the engineering objective and intent the engineer described: a hydrogen fuel-cell interface across all seven domains, concentrated four-species transport model because oxygen in nitrogen is not dilute, and anhydrous proton conduction because PA-doped PBI carries charge by a Grotthuss mechanism rather than water drag. The engineer supplied the benchmark parameters and confirmed each call before the solve.

3D swept-hex production mesh · 17,696 elements · cathode CL 10 divisions

Mapped mesh · X–Z cross-section

Setting	Value	Completed by
Physics interfaces	Hydrogen Fuel Cell (PEM): anode flow channel, anode GDL, anode GDE, membrane, cathode GDE, cathode GDL, cathode flow channel	ARGUS
Species transport	Concentrated, H₂, O₂, H₂O, N₂	ARGUS
Membrane conduction	Anhydrous (Grotthuss); no electro-osmotic swelling	ARGUS
Flow	Free & Porous Media; Brinkman in GDL and electrodes	ARGUS
Dimensionality	3D, for in-plane (X–Y) current across the rib	ENGINEER
Solver	Five-step staged, continuation 0.95 → 0.40 V, PARDISO	ARGUS
Parameters & mesh	Benchmark values; per-layer graded, engineer-verified	ENGINEER

Argus called the physics correctly

Before consulting any reference simulation, Argus established that PA-doped PBI conducts protons anhydrously, ruled out two-phase flow because water remains vapour at 180°C, and specified concentrated species transport. These are the calls that decide whether the underlying physics is even the right one.

The engineer stayed in the loop

The engineer set the targets, made the decision to prepare a 3D simulation in order to account for in-plane current along the rib, supplied the benchmark parameters and signed off on each of Argus’s calls before the solve. Validation stayed with the engineer, who validated the converged result.

Validation & Results

The benchmark is COMSOL’s published ht_pem simulation, based on the work of Ubong, Shi and Wang (2009). To keep the comparison honest, that benchmark was first re-solved from scratch by Argus on its own and reproduced to 0.00% at every point, establishing the baseline, and only then was the agent’s own coupled model under test measured against it.

Cell voltage (V)	Benchmark (A/cm²)	This model (A/cm²)	Deviation
0.95	0.00160	0.00160	0.00%
0.80	0.05043	0.05040	0.06%
0.70	0.20203	0.20181	0.11%
0.60	0.44154	0.44095	0.13%
0.50	0.72223	0.72134	0.12%
0.40	1.00392	1.00343	0.05%

1 · The cathode starves first

Oxygen concentration falls from 0.17 to 0.03 (mole fraction) down the cathode.

2 · Water pools at the cathode

The cathode runs far wetter than the anode, about 0.32 against 0.16 (water mole fraction).

3 · What the 3D model shows

Ionic current falls from 11.3 to 8.4 kA/m² and varies across the rib.

4 · Proof the transport is live

The straight line confirms reactant consumption and product formation are genuinely coupled to the flow.

Agent Design Input

This build tested Argus’s reasoning as much as its output. Argus held the physics that stood up against a published benchmark. The simulation carried configuration errors that passed every clean solve. Argus surfaced several errors and the engineer surfaced the rest. Here is the honest read.

Takeaway

The agent built its model first, independently from the physics description and parameter set, without looking at how the benchmark was constructed. That model ran cleanly but came out ~38× too high at open circuit voltage. Argus noticed the tell: cathode oxygen concentration sitting at 1.5 times the benchmark’s for the same mole fraction and pressure. Argus applied the ideal gas law to back out a model temperature near 300 K instead of the expected 453 K. Argus traced this fault back to gas properties evaluated at room temperature. The cause was one missing setting: the shared temperature input was never applied, so the gas properties defaulted to room temperature. Argus reached that from the result data alone.

One pass, every node

Where the work is scripted and systematic, Argus is fast and exact. Argus applies a single property across every node in one reproducible, logged pass, runs a full differential comparison of two models to pin down a defect, and lays down the staged solver configuration in a single go. This is the repetitive GUI work that is prone to human error. Argus completes this work the same way every time, with every step logged and traceable.

Argus flagged what others missed

A supplied expression had added a constant to a temperature difference in kelvin, a dimensional error. Argus caught it before building the simulation and, rather than silently substituting a fix, flagged it, entered the expression as instructed, and reported the result.

How Argus read the physics

Before seeing the benchmark, Argus identified the appropriate HT-PEM physics, ruled out the low-temperature machinery that does not apply, and proposed a continuation-in-voltage solver strategy that was structurally identical to the benchmark’s own.

→

Next steps and Argus

The next step is to take the validated model off its single operating point: thermal coupling for a non-isothermal cell, a humidification sweep to map the shutdown-condensation boundary, and a transient cooldown for the acid-leaching risk. The setup already exists, so Argus re-runs each new condition in minutes rather than rebuilding the entire simulation study. That turns a single validated result into a simulation that can explore the cell’s full operating envelope.

Think through > click through.
Let Argus handle the rest.

Your AI agent for engineering workflows

Want to try Argus on your own multiphysics workflow?

Visit www.cosmon.com to start a trial.