About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Materials Processing and Fundamental Understanding Based on Machine Learning and Data Informatics
|
| Presentation Title |
Denoising Diffusion Probabilistic Model for Data Augmentation and Inverse Design of Structural Materials |
| Author(s) |
Yoon Suk Choi, Libin Zhang, Taejoo Lee, Sujeong Kim |
| On-Site Speaker (Planned) |
Yoon Suk Choi |
| Abstract Scope |
Three case studies explore the potential of denoising diffusion probabilistic model (DDPM) as a tool for the data augmentation and inverse design of structural materials. In the first case, DDPM was adopted to generate a synthetic composition-ultimate tensile strength (UTS) dataset to search compositional pools of 7xxx-series aluminum alloys with the desired UTS. The DDPM-generated composition-UTS dataset screened by machine learning models reasonably captured aluminum alloy pools with desired UTS ranges. In the second case, the DDPM was employed for the data augmentation, and the compositional design and optimization in developing refractory high entropy alloys (HEAs). A DDPM-assisted generative inverse design framework was proposed, and its efficient compositional optimization was demonstrated. In the last case, DDPM was implemented for the short-term creep data-based log-term creep life prediction. A synthetic creep dataset was generated by DDPM using physics-based predictions of short-term creep data, and its long-term creep life predictability was assessed. |