Case Study: Using LMS Amesim for Fault Diagnosis of a Commercial PEM Fuel Cell System

A peer-reviewed paper published in Advances in Science, Technology and Engineering Systems Journal ("The Use of LMS AMESim in the Fault Diagnosis of a Commercial PEM Fuel Cell System") documents a practical, resource-constrained modeling workflow worth noting for anyone doing energy-system diagnostics work in Amesim.

Methodology and a practical constraint

The researchers modeled a commercial ElectraGen 3 kW PEMFC system using Siemens LMS AMESim 14. Rather than deriving the fuel cell physics from first-principles equations, they used Amesim's embedded fuel cell component library, which already encodes the standard governing equations — reducing the modeling effort to parameterization rather than equation derivation.

Practical/resource lesson: parameter identification was done via manual trial-and-error matching against experimental data, explicitly not using genetic-algorithm-based optimization — because the team did not have access to multiple software licenses needed to run that optimization approach in parallel. This is a real-world constraint worth flagging: Amesim's optimization tooling assumes license availability that not every team has, and manual calibration remains a valid (if slower) fallback.

Faults simulated

The model was used to simulate five distinct fault conditions by perturbing specific model inputs:

1. Membrane drying — reduced humidifier target to 10%.
2. Membrane flooding — increased humidity to 100% while lowering stack temperature.
3. Air leakage — introduced a relief valve with a 10 g/s leak rate.
4. Hydrogen leakage — added a relief valve after the hydrogen supply.
5. Cooling system failure — elevated stack temperature to 75°C.

Diagnostic method and a key practical finding

The diagnosis approach compared actual system performance against the simulated model's predictions using five residual-based diagnostic signals tracking stack voltage, current, temperature, and pressures.

The paper explicitly notes a limitation of the naive approach: "the diagnosis approach based on instantaneous residuals... is impractical and highly sensitive to electromagnetic disturbance pulses." The fix that worked: switching to a moving 10-second window calculation of the residuals rather than instantaneous sampling. This windowed approach successfully detected and isolated all five simulated faults with minimal false positives.

Practical takeaways for KB users

• When manufacturer data/parameters are confidential or unavailable, Amesim's pre-built component libraries (encoding standard governing equations) let you build a credible model without deriving physics from scratch — parameterization becomes the main task.
• Manual trial-and-error calibration against experimental data is a legitimate fallback when optimization tooling (genetic algorithms, parallel license pools) isn't available — don't assume you need HEEDS or Amesim's built-in GA optimizer for every calibration task.
• If you're building a model-based fault detection/diagnosis pipeline, be aware that raw instantaneous residuals are noise-sensitive (especially to electromagnetic interference in real sensor data) — a moving-window averaging approach is a proven, simple robustness improvement.

Source: The Use of LMS AMESim in the Fault Diagnosis of a Commercial PEM Fuel Cell System – ASTESJ