South Korean researchers develop AI framework to optimize fuel cells
In South Korea, researchers from Korea have now developed an artificial intelligence (AI)-guided optimization framework that dramatically reduces the computational effort needed to optimize solid oxide electrolysis cell (SOEC) operation. By combining high-fidelity computational fluid dynamics (CFD) simulations with an AI-driven active learning framework, the researchers rapidly identified operating conditions that improve hydrogen production efficiency while maintaining the thermal stability required for long-term operation. This paper was made available online on 25 June 2026 and has been published in Volume 303, Part 1, of the journal Applied Thermal Engineering on August 1 2026.
Instead of evaluating every possible operating condition, the AI framework learns from each completed simulation and predicts which operating conditions are most likely to provide valuable new information. This allows researchers to focus computational resources on the most promising operating regions, replacing exhaustive trial-and-error searches with a faster, more data-efficient optimization strategy.
Unlike conventional approaches that seek a single optimum, the framework identifies a Pareto-optimal operating region that balances two competing objectives: maximizing electrochemical performance while minimizing temperature differences that can accelerate material degradation and reduce device lifespan. This gives engineers the flexibility to select operating conditions based on practical priorities, whether maximizing efficiency, improving durability, or achieving the best compromise between the two.
The framework improved the electrochemical performance index (EPI) by 14% while reducing in-plane temperature differences by 80% compared with the baseline operating condition. Compared with conventional random sampling using the same computational budget, the AI-guided approach achieved 2.5% higher final EPI and a 90.5% lower final temperature difference.
Most significantly, the framework achieved these results using only 17 high-fidelity CFD simulations. An exhaustive search across the same operating space would have required 6,561 simulations, equivalent to approximately 22,963.5 computational hours. The AI-guided framework achieved comparable optimization performance in just 60 hours, demonstrating its potential to shorten engineering design cycles and accelerate the development and commercialization of efficient green hydrogen technologies.
Category: Research
