HYDRA: Hybrid Multimodal AI-driven Design for Robust High-Entropy Oxide Electrocatalysts in ORR

The oxygen reduction reaction (ORR) is critical for electrochemical energy devices, yet its slow kinetics and the high cost of platinum-group metal catalysts necessitate alternative materials. High-entropy oxides (HEOs) show immense potential, but their vast design space challenges traditional discovery. This study introduces HYDRA, a Hybrid Multimodal AI-driven Design platform, to accelerate HEO catalyst discovery through intelligent design and cross-modal validation. HYDRA comprises four synergistic modules: a multi-graph neural network for accurate ORR performance prediction, a large language model for extracting synthesizability knowledge and generating protocols, a multi-objective evolutionary algorithm for optimization and design insights, and a neural network for microstructural validation. Our results demonstrate that HY-DRA's predictive model significantly outperforms state-of-the-art baselines in estimating ORR half-wave potential and stability. Ablation studies confirm the critical contributions of multi-scale structural information and defect descriptors. The large language model generates highly feasible and novel synthesis protocols. A successful end-to-end case study, where a HYDRA-designed HEO was synthesized and experimentally validated with high agreement, coupled with precise microstructural confirmation, underscores the framework's robustness. HYDRA thus provides a powerful, interpretable, and practically validated approach to overcome conventional HEO catalyst discovery limitations, accelerating the development of next-generation ORR electrocatalysts.