Efficient design of hydrogen-resistant ultra-high-strength steels via active learning and multiscale characterization

• An active learning strategy is developed to efficiently design steel. • VC precipitates act as hydrogen capture sites to reduce HE sensitivity. • Pareto front analysis is optimized for multi-objective optimization. High-strength martensitic steels are extensively used in aerospace, automotive manufacturing, marine engineering, and other demanding applications due to their exceptional mechanical properties. In this work, an active learning strategy integrated with a genetic algorithm is proposed to accelerate the efficient design of ultra-high-strength steels (UHSS) with enhanced resistance to hydrogen embrittlement (HE). After three iterative optimization cycles, alloy compositions meeting the targeted performance metrics are successfully identified. Mechanical properties of selected alloys are subsequently validated through hydrogen-charged slow strain rate tensile (SSRT) testing. Concurrently, high-resolution transmission electron microscopy (HRTEM) and atom probe tomography (APT) are employed to elucidate key microstructural features. According to HRTEM statistical results, volume fraction of VC precipitates is about 0.623%, which acts as a hydrogen capture site and can effectively reduce sensitivity to HE. This methodology not only streamlines design process for UHSS and markedly enhances research and development efficiency, but also provides a solid theoretical foundation for performance optimization and practical application of high-strength steels.