Artificial intelligence-driven unraveling of the critical factor of heteroatom-doped carbon-based ORR catalysts

Abstract

This work leverages artificial intelligence (AI) to explore the crucial factor in optimizing heteroatom-doped carbon-based ORR catalysts. A dataset of 3375 data points, 23 features, and 2 labels is analyzed using 8 AI algorithms, achieving over 72% accuracy. AI findings reveal graphitic-N content as the primary factor affecting the half-wave potential. Moreover, the pyrolysis holding time (23.8 to 27.2%) contributes more significantly to graphitic-N content than the pyrolysis temperature (13.3 to 14.7%), challenging prior studies that prioritize temperature as the main factor. AI uncovers a novel empirical observation, revealing that the pyrolysis holding time influences the half-wave potential by modulating graphitic-N content. To validate these findings, graphitic-N-doped carbon-based ORR catalysts are synthesized using urea and XC-72. A strong positive correlation between graphitic-N content and half-wave potential is confirmed, with a 1 h extension of pyrolysis time, leading to a more significant increase in graphitic-N content (0.25 to 0.60 at%) than a 100 °C increase in temperature (0.12 to 0.32 at%). AIMD and DFT calculations show that graphitic-N atoms are more thermally stable than pyrrolic-N and pyridinic-N, which explains the increase in graphitic-N content with longer pyrolysis times, while pyrrolic-N and pyridinic-N atoms volatilize. This work presents an innovative AI-driven experimental paradigm for ORR catalyst design.