Deep learning-guided multi-material optimization of Al2O3 washcoated additively manufactured monoliths for efficient ethanol steam reforming and hydrogen production

Hydrogen is a promising clean energy carrier for achieving carbon neutrality, while ethanol steam reforming (ESR) provides a sustainable route for renewable hydrogen production due to the availability of bioethanol and its high hydrogen yield. Additively manufactured metallic monoliths have gained attention as catalyst supports because of their excellent heat transfer, low pressure drops, and design flexibility. However, catalyst performance largely depends on the adhesion and thermal stability of the Al2O3 washcoat, making conventional optimization expensive and time-consuming. This study presents a deep learning-assisted multi-material optimization framework for Al2O3 washcoated monoliths fabricated by selective laser melting using 304L, 316L, and 310S stainless steels and Inconel 625. Comprehensive material characterization, including scanning electron microscopy, energy-dispersive x-ray spectroscopy, x-ray diffraction, Brunauer–Emmett–Teller, surface roughness, and adhesion testing, was integrated with catalytic evaluation under ESR conditions. A hybrid CNN–transformer model was developed to predict washcoat adhesion, thermal stability, ethanol conversion, hydrogen yield, and catalyst durability. The proposed model achieved an R2 value of 0.997, an MAE of 1.28, and a root mean square error of 1.84. The optimized Inconel 625 monolith exhibited 99% ethanol conversion, 91% hydrogen yield, and retained 94% catalytic activity after 200 h of continuous operation. These results demonstrate that integrating additive manufacturing with explainable deep learning provides an efficient and scalable approach for optimizing structured catalysts and accelerating intelligent hydrogen production.