(159a) Machine Learning Technique for Core Materials Design Toward Active Systems in Energy Storage and Conversion Reactions

Nanoparticles are core materials for various electrochemical reactions for energy storage and conversion. Since last several decades the overall systematic efficiency utilizing the reactions have not reached to the level of our target due to several unsolved issues caused by nano-materials' inert functionality, high cost and instability over long-term operation.

In atomic-level computational electrochemistry a new research paradigm for desired materials design has been established through consistent and tight combination of IT-based artificial intelligence (AI) technology and machine learning algorithm standing on supercomputer architectures. This presentation demonstrates the computational methodologies of high-throughput screening of promising nanoparticle candidates for electrochemical energy conversion and storage systems. Using the first-principles density functional theory calculations and frontier realtime and atomic level experimental measurements we acquire reliable and accurate materials properties as input to activate machine learning model. To elevate the accuracy even to higher level we incorporate active-learning and multi-fidelity methods.

It is shown that an AI-based neural-network model is very useful for identifying multi-component electrocatalysts toward three reactions (HER, OER, ORR) at the same time. A computational platform making the pipeline automatic is demonstrated.