Seminar & Colloquium
[콜로퀴엄: 12월 4일(수), 오후 5시] 부산대학교 재료공학부, 최윤석 교수
Title
알면 쓸모있는 구조재료 인공지능 기법들
(Some useful artificial intelligence approaches for structural materials)
Speaker
부산대학교 재료공학부, 최윤석 교수
Education
- 2001.05. Department of Materials Science & Engineering, Carnegie Mellon University, 공학박사
- 1998.05. Department of Materials Science & Engineering, Carnegie Mellon University, 이학석사(Coursework Maters)
- 1996. 02. 부산대학교 공과대학 금속공학과, 공학석사
- 1994. 02. 부산대학교 공과대학 금속공학과, 공학사
Experience
- 2018.03 – 현재 부산대학교 공과대학 재료공학부, 교수
- 2021.01 – 2023.12 대한금속‧재료학회 학술이사
- 2020.01 – 2021.12 Metals and Materials International, 편집위원
- 2020.01 – 2020.12 대한금속‧재료학회 타이타늄분과위원회 위원장
- 2016.01 – 2018.12 대한금속‧재료학회 국제담당이사
- 2018.03 – 2020.03 부산대학교 공과대학 기획부학장
- 2013.03 - 2018.02 부산대학교 공과대학 재료공학부, 부교수
- 2002.03 - 2013.02 UES, Inc., Dayton, OH, USA, 책임연구원, (미 공군연구소 계약 연구원)
- 2001.05 - 2002.01 Department of Civil & Environmental Eng., Carnegie Mellon University 박사후과정(Strain-Gradient Plasticity ABAQUSTM UMAT 코드 개발)
| Date | Wednesday, December 4th, 2024
| Time | 17:00 ~
| Venue | 43동 101호
[Abstract]
Several artificial intelligence (AI) techniques were explored as a tool for data preprocessing, forward machine learning, data augmentation and inverse design of structural metallic materials. First, data preprocessing and machine learning procedure were generalized for structural metallic materials data, typically consisting of composition, processing conditions and mechanical properties. Second, a machine learning-based data preprocessing methodology was proposed to eliminate outliers (or noises) for the creep data collected from various literature survey. A second-generation Ni-base single crystal superalloy (CMSX-4), was chosen as a target material for the creep data collection, since its creep data were relatively readily available from a variety of previous studies. Third, denoising diffusion probabilistic model (DDPM) was assessed as a potential tool for the data augmentation and inverse design of structural metallic materials. A DDPM-assisted generative inverse design framework was proposed, and its efficient compositional optimization was demonstrated. Lastly, transfer learning and imbalance learning techniques were explored as a tool for handling small, but still meaningful, materials datasets.
| Host | 한흥남 교수(02-880-9240)