Artificial Intelligence and Expert Cooperative Design of Non-isothermal Aging Heat Treatment Schedules for Improving 0.2% Proof Stress in γ – γ' Binary Ni-Al Alloys V. Nandal1, S. Dieb1, D. S. Bulgarevich2, T. Osada2, T. Koyama3, S. Minamoto1, M. Demura1 1 Center for Basic Research on Materials, National Institute for Materials Science (NIMS) 2 Research Center for Structural Materials, National Institute for Materials Science (NIMS) 3 Department of Materials Design Innovation Engineering, Nagoya University In this study, we devised a unique aging heat treatment method to increase the high-temperature strength by utilizing state-of-the-art Artificial Intelligence (AI) algorithms in binary Ni-Al alloys. We challenged ourselves to find a non-isothermal aging (NIA) pattern that surpasses isothermal aging from a huge combination of complex aging heat treatment patterns, including increasing temperature and lowering temperature (about 3.5 billion possible ways) while isothermal aging was used as a traditional method. As a result, we succeeded in finding 110 outperformed NIA heat treatment schedules from 1620 trials. In addition, through the analysis of the top 5 NIA patterns discovered by AI, we have derived the idea of a new two-stage aging system that combines high-temperature for a short time and low-temperature for a long time and confirmed that this surpasses the search results of AI. These results suggest that AI and experts can jointly develop new process methods. Keywords: Machine learning, Artificial Intelligence, Monte Carlo Tree Search Highlier Efficient Neural Network Interatomic Potential of α-iron and Hydrogen System Shihao Zhang 1, Fanshun Meng 1 and Shigenobu Ogata 1 1 Department of Mechanical Science and Bioengineering, Graduate School of Engineering Science, Osaka University, Osaka, 560-8531, Japan Artificial neural network potential (NNP) provides an accurate tool to describe the atomic interactions toward atomic-scale understanding of hydrogen embrittlement in α-iron. However, the current NNP still suffers the high computational cost comparing with empirical potential, which much limits its application for many problems of practical interests in hydrogen embrittlement. In this work, following our previous work of iron-hydrogen NNP [Physical Review Materials 5, 113606 (2021)], a new NNP was developed and validated, which not only quantitatively describes the atomistic details of hydrogen behavior in the defective α-iron system with the accuracy of density functional theory, but also shows much higher efficiency (∼40 times faster) than the previous NNP. Furthermore, the NNP was applied to study the hydrogen embrittlement using the polycrystalline model and sharp penny-shaped crack model of α-iron. We expect that this NNP would provide a high-efficiency tool for atomic-scale understanding of hydrogen embrittlement. 76PP44--1155 PP44--1166 Poster Presentation ｜NIMS Award Symposium 2023 P4 ｜ Modeling, Aging, Ni-based alloys.
元のページ ../index.html#76