Data-driven multi-objective optimization of performance and thermal stability in direct ammonia solid oxide fuel cells

Direct ammonia solid oxide fuel cells (DA-SOFCs) represent a highly efficient and sustainable energy conversion technology. Nevertheless, the ‘trade-off’ between the performance and thermal stability remains challenging. Multi-physics simulation (MPS) was utilized in this study to gain insight into the effects of key operational and structural parameters on the output power density (Pout), electrical efficiency (ηcell), and maximum temperature gradient (GT,max) at the cell level. It was revealed that enhancements in performance via parameter optimization are often accompanied by a substantial increase in GT,max, thereby exacerbating the risk of thermomechanical failure. To address this point, a multi-objective optimization framework integrating MPS, artificial neural networks (ANN) and genetic algorithms (GA) was developed in this study to break the ‘trade-off’. The ANN surrogate model was trained, validated and tested using 3125 samples generated by MPS, achieving a coefficient of determination (R2) values exceeding 0.995 and significantly reducing the computation time for evaluating these samples from 360 h to just 0.223 s. The ANN surrogate model was then integrated with a GA to perform multi-objective co-optimization. Under the strict constraint of GT,max ≤ 10 Kcm‐1, the multi-objective optimization framework achieved 102.8% and 27.4% enhancement in Pout and ηcell, respectively. These results demonstrate that the framework is capable of synergistically optimizing the performance and thermal reliability of DA-SOFCs and tailoring the operation condition with low computation cost and high efficiency.