Speakers
Description
The shift toward cleaner energy systems has placed hydrogen at the forefront of fuel research, particularly for distributed power generation where Micro Gas Turbines (MGTs) offer a practical and flexible solution. Unlike conventional fuels, hydrogen produces no CO₂ at the point of combustion, and even a 75% hydrogen blend can cut emissions by up to 50% compared to natural gas. A major roadblock is the lack of real-world hardware failure data for these new systems. To address this, a physics-informed framework is proposed that combines a validated recuperated Brayton cycle thermodynamic model with a Random Forest classifier for fault identification. The baseline thermodynamic model is validated against experimental data from a hydrogen-fuelled MGT test rig, then used to generate synthetic fault datasets by simulating degradation in key components across a range of operating conditions. A Random Forest classifier is trained on this simulated data to identify fault types using only standard measurable parameters. The classifier successfully identifies all simulated faults providing a strong proof-of-concept. The study demonstrates that using physic-based simulations to train machine learning models is a practical way to bypass data scarcity in early-stage hydrogen systems, though field application will eventually require testing against real mechanical failures.
| Topic Area | Smart Operations and Maintenance |
|---|