Machine Learning-Guided Surface Strain Engineering in Connected Platinum-Nickel Nanoparticle Catalysts for Advanced Oxygen Reduction Performance.

Engineering the surface structure of catalysts is critical for achieving high intrinsic activity in the oxygen reduction reaction (ORR). We report a machine-learning (ML)-guided materials design strategy for the synthesis of support-free, connected nanoparticle catalysts with enhanced activity. ML analysis of a dataset comprising 210 Pt-based ORR catalysts quantitatively evaluated the relative importance of multiple structural, compositional, and electronic descriptors, identifying surface compressive strain (≈-4%) as an effective integrated descriptor strongly associated with ORR specific activity (SA). Guided by this insight, a H2-annealing-induced surface structuring approach was developed to construct an interconnected porous Pt-Ni nanoarchitecture with a Pt-skin (≈3 atomic layers) and tunable compressive strain (≈-3%) over a Ni-enriched subsurface. The optimized catalyst synthesized under 100% H2-annealing exhibits exceptional ORR activity, with a SA of 5.1 ± 0.5 mA cmPt - 2, corresponding to a 12-fold enhancement relative to commercial Pt/C. A linear correlation between surface strain and SA experimentally validates ML predictions and highlights the importance of strain-engineered surfaces in electrocatalysis. Furthermore, the connected nanonetwork demonstrates remarkable electrochemical durability, retaining substantial compressive strain and exhibiting minimal Ni dissolution after 10,000 potential load cycles. This work establishes a generalizable materials design framework for developing next-generation high-activity, durable electrocatalysts for energy conversion technologies.