AI cuts solid-oxide cell optimization from 6,561 runs to 17

Seventeen simulations replaced a search that would have demanded 6,561 in the Korean researchers’ workflow for optimizing a solid oxide electrolysis cell, or SOEC. The team at Seoul National University of Science and Technology used artificial intelligence to decide where to look next rather than asking a computer to test every operating condition.

The method combines high-fidelity computational fluid dynamics, simulations that model how fluids, heat and related physical effects behave, with active learning. After each completed simulation, the AI predicts which untested condition is most likely to add valuable information. That concentrates computing power on promising regions instead of spreading it across an exhaustive trial-and-error search.

The target is not a single perfect setting. The framework maps a Pareto-optimal operating region, balancing two competing goals: improving electrochemical performance while limiting temperature differences inside the cell. Large differences can accelerate material degradation and shorten device life, so engineers can choose a point according to whether efficiency, durability or a compromise matters most.

The reported gains are substantial within the simulation study. The electrochemical performance index rose by 14% against the baseline operating condition, while in-plane temperature differences fell by 80%. Using the same computational budget, the AI-guided approach produced a 2.5% higher final EPI and a 90.5% lower final temperature difference than conventional random sampling.

So what changes in practice? Engineers could explore promising designs and operating strategies sooner, using 60 hours of computation rather than the approximately 22,963.5 computational hours associated with an exhaustive search over the same space. That could shorten development cycles for green hydrogen technologies. The boundary is clear: these are high-fidelity simulations reported by the research team, not evidence that an SOEC has already been improved in commercial operation. The researchers say the same strategy could also be applied to fuel cells, batteries and catalytic systems where simulations are expensive.