| About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Impurity-Tolerant Alloy Design, Development, and Production
|
| Presentation Title |
Generative-AI for Impurity-Tolerant Robust Alloy Design |
| Author(s) |
Patxi Fernandez-Zelaia, Saket Thapliyal, Rangasayee Kannan, Peeyush Nandwana, Yukinori Yamamoto, Andrzej Nycz, Vincent Paquit, Michael Kirka |
| On-Site Speaker (Planned) |
Patxi Fernandez-Zelaia |
| Abstract Scope |
Inverse materials design is essential in many energy related applications. The optimization problem is made difficult by non-linear performance-composition relationships and potential solution non-uniqueness. Furthermore, existing physics-based tools only operate in the forward direction; a user inputs a chemistry and obtains a prediction. Denoising diffusion probabilistic models (DDPMs) are a new class of generative models that have been shown to be extremely expressive across various data modalities e.g. images, text, audio, tables, etc.. In this talk we present recent work using DDPMs for an impurity-tolerant robust alloy design for binder jet additive manufacturing. Results indicate that the established model is extremely flexible and well suited for various optimization problems. Furthermore, due to their probabilistic nature these models are well suited for uncertainty quantification. We envision that future alloy design workflows will extensively use these DDPMs as advanced “search” tools bolstering the value of both experimental and computational approaches. |