http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
기계학습을 활용한 2차원 이종접합 헤테로구조의 수소발생 성능 예측
김은송(Eunsong Kim),팜 티 후에(Thi Hue Pham),민경민(Kyoungmin Min),신영한(Young-Han Shin) 대한기계학회 2023 대한기계학회 춘추학술대회 Vol.2023 No.11
While efficient and cost-effective catalysts are needed for long-term hydrogen production, low-dimensional interfacial approaches have been developed to increase catalytic activity performance in hydrogen evolution reaction (HER). We calculated the Gibbs free energy change (ΔGH) in hydrogen adsorption in the two-dimensional lateral heterostructures (LHS) at various adsorption points in each unit-cell using density functional theory (DFT). We developed three types of descriptors (position feature, weight feature, average feature) that may be utilized universally in 2D materials, also can explain ΔGH based on different adsorption sites in a single LHS combining basic LHS information (the type and quantity of neighboring atoms around the adsorption point). Furthermore, we trained machine learning (ML) models with the specified descriptors to predict the potential conjunction and adsorption sites within the LHS for HER catalysts. ML model in this research reached R2 score of 0.95. Additionally, 8 LHS materials were examined successfully with the ML model.
다목적 최적화 방법을 통한 유망한 물질의 신속한 발견을 위한 능동적 학습 과정 검증
김태현(Taehyun Kim),김민선(Minseon Kim),김은송(Eunsong Kim),홍은화(Eunhwa Hong),민경민(Kyoungmin Min) 대한기계학회 2022 대한기계학회 춘추학술대회 Vol.2022 No.11
In the field of materials science, research is actively underway to search for materials that satisfy certain target properties. Although consideration is simple when there is one target property, it is difficult to consider in the case of multi target properties, which is most of the actual reality problems. Therefore, there is a need for a multi-objective optimization (MOO) algorithm with good performance that can consider several targets at the same time. In this study, we verify the performance of the active learning process through the MOO methods by applying the MOO methods to two-dimensional materials database. In the future, verification results for another database will be obtained, and verification results will be expanded to general materials. Through performance verification, efficient MOO guidelines can be presented.
다목적 최적화 방법을 통한 유망한 물질의 신속한 발견을 위한 능동적 학습 과정 검증
김태현(Taehyun Kim),김민선(Minseon Kim),김은송(Eunsong Kim),홍은화(Eunhwa Hong),민경민(Kyoungmin Min) 대한기계학회 2022 대한기계학회 춘추학술대회 Vol.2022 No.11
In the field of materials science, research is actively underway to search for materials that satisfy certain target properties. Although consideration is simple when there is one target property, it is difficult to consider in the case of multi target properties, which is most of the actual reality problems. Therefore, there is a need for a multi-objective optimization (MOO) algorithm with good performance that can consider several targets at the same time. In this study, we verify the performance of the active learning process through the MOO methods by applying the MOO methods to two-dimensional materials database. In the future, verification results for another database will be obtained, and verification results will be expanded to general materials. Through performance verification, efficient MOO guidelines can be presented.
머신 러닝을 활용한 고엔트로피 가넷 구조의 전고체 전지를 위한 고체 전해질 발견
선지원(Jiwon Sun),김은송(Eunsong Kim),김주오(Juo Kim),김준철(Joonchul Kim),민경민(Kyoungmin Min) 대한기계학회 2023 대한기계학회 춘추학술대회 Vol.2023 No.11
Research in all-solid-state batteries (ASSBs) has surged due to their improved energy density and safety over liquid-based lithium-ion batteries. To enhance electrochemical performance, stable solid-state electrolytes (SSEs) are crucial. This study employed a novel machine learning (ML) screening platform to explore 161,280 high-entropy (HE) garnet-type SSE originated from known Li₇La₃Zr₂O<SUB>12</SUB> (LLZO) structures. Initially, an ML-based surrogate model identified electronconductive (bandgap < 1 eV) and thermodynamically unfavorable (energy above hull > 0.035 eV) materials to prevent short-circuits and decomposition. A recently developed ML interatomic potential (M3GNet) guided atomic arrangement for stability. Elastic properties were predicted to ensure dendrite suppression and interfacial stability. Molecular dynamics using M3GNet confirmed Li diffusivity. In conclusion, 23 promising HE garnet materials were identified for advanced ASSBs with these methods.