脱炭素エネルギー先導人材育成フェローシップ 2022年度
25/32

wolle機械学習は研究開発を加速させる強力なアプローチとして材料科学の世界をも席巻しています。Kan君は機械学習的を使ったアニオン交換膜用高分子の分子設計探索手法の開発に着手し、これまで研究者の知識と経験に頼ってきた分子設計の加速にチャレンジしています。Kan君には、この研究を踏み台に、難易度の高いとされる高分子材料の機械学習の可能性を広げて欲しいと考えています。さらに、多様な分野の人材が集うこの脱炭素フェローシップでの活動を通して視野を広げ、幅広い分野においてその才能を発揮して欲しいと願っています。Q-Energy Innovator Fellowship23工学府応用化学専攻 博士後期課程1年工学研究院 教授The largest story of 21st century – climate change, can be tack-led by drastically reducing the emission of greenhouse gases. Fuel cells, such as anion exchange membrane fuel cell (AEMFC), which operate on hydrogen gas and emit only water as byproduct, are one such option for decarbonization. Unfortunately, AEMFC is not ready for widespread adoption due to the low ion conductiv-ity and durability of its anion exchange membrane (AEM), which is a key component for ion transportation between cathode and anode in AEMFC. Research and development (R&D) of the im-provement of AEM has mostly focused on experiment-centric ef-forts, which require time, money, and human resource. Materials informatics (MI), which utilizes machine learning (ML) model, plays a key role in speeding up the R&D cycle of AEMFC for the early achievement of decarbonized society. This is because ML models enable high-throughput materials exploration as well as reduction in the number of syntheses that need to be physically performed. However, the difficulty of interpreting the prediction logic of ML model, often referred to as a black box, presents a barrier to the widespread implementation of ML as a research methodology in the AEMFC field. This study aims to establish a machine learning model that can be used to predict the target properties of AEM, with prediction logic interpretable and comprehensible by human researchers. Conventional methods focus on mathematically representing chemical structures using numbers, resulting in hard-to-decipher results. Instead, we are working on using molecular graphs to rep-resent the chemical structure of AEM. This graphical representa-tion is more intuitive, and helps researchers to better understand what makes a better AEM. Researchers can then use this knowl-edge to design new AEM polymer structure in the lab without having to go through rounds of syntheses to discover new poly-mer structures, thereby accelerating the R&D of AEMFC.指導教員からメッセージParticipation of polymer science conference held in Hokkaido, Japan, 2022.藤ヶ谷 剛彦Establishing Machine Learning Model for use in Anion Exchange Membrane with Prediction Logic Comprehensible by Humanアニオン伝導膜形燃料電池の材料探索における化学的解釈可能な機械学習の開発Phua Yin Kan18f

元のページ  ../index.html#25

このブックを見る