| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI/Data Informatics: Computational Model Development, Verification, Validation, and Uncertainty Quantification
|
| Presentation Title |
Novel Superhard Materials Synthesis Using Generative ML |
| Author(s) |
Lavanya Mohnani, Taylor Sparks, Michael Alverson |
| On-Site Speaker (Planned) |
Lavanya Mohnani |
| Abstract Scope |
Superhard materials, such as diamond, hold immense potential for all disciplines, but due to cost and difficulty of creation, their advantageous properties are rarely utilized. This work aims to present a novel computational method for the synthesis of new superhard materials using a Wasserstein-enhanced generative adversarial network (WGAN). Following training of the GAN on CIFs from Pearson’s Crystal Database, a library known as M3GNet, which can mathematically predict the energy above the hull of any given structure will be used as a filter to eliminate unstable crystals, a novel technique for GAN stability. Additionally, the CHGNet neural network will be mathematically incorporated into the generator's loss function (higher instability would correspond to higher loss), ensuring only stable superhard crystals are created by the GAN following convergence. This study presents several potential techniques for state-of-the-art ML performance, propelling discovery for both computational science and materials discovery. |
| Proceedings Inclusion? |
Planned: |