About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Aluminum Alloys: Development and Fabrication Symposium and Supplier Forum
|
| Presentation Title |
AI-Driven Closed-Loop Alloy Design for High-Performance Aluminum Alloys through Semi-Supervised Active Learning |
| Author(s) |
Kwangjun Euh, Hyeon-Woo Son, Uijae Lee, Yong Hee Jo, Sang-Ik Lee, Kisub Cho |
| On-Site Speaker (Planned) |
Kwangjun Euh |
| Abstract Scope |
Conventional aluminum alloy development has relied on empirical, single-property-focused approaches, resulting in substantial time and cost, while existing materials data services remain concentrated on commercial alloy systems, leaving significant data gaps. This study presents an AI-driven closed-loop alloy design framework integrating a composition–process–microstructure–property database, semi-supervised autoencoder-based active learning, high-throughput experiments, and computational materials engineering. The proposed latent-space embedding significantly improved Gaussian Process Regression prediction by effectively utilizing labeled, pseudo-labeled, and unlabeled data. Through five iterative query–synthesis feedback cycles, ten candidate high-strength aluminum alloys were identified and fabricated at bench scale. The resulting alloy achieved a yield strength of 638 MPa, tensile strength of 672 MPa, specific strength of 244 MPa/(g·cm⁻³), and fatigue strength of 200 MPa (10⁷ cycles). |
| Proceedings Inclusion? |
Planned: Light Metals |
| Keywords |
Aluminum, Machine Learning, Mechanical Properties |