| Abstract Scope |
Local chemical ordering (LCO) is often a key factor in a material’s damage tolerance. For example, transmutation helium in structural metals for fusion reactors can shorten the metals’ service life from ~10 years to <6 months, depending on its arrangement within the metal. However, atomic-scale LCOs are difficult to predict and characterize, making materials design challenging. In this talk, I will discuss how a synergistic combination of computational, experimental, and data-driven approaches can help engineer LCOs. First, I will focus on helium ordering in nuclear materials, where we computationally designed preferential helium sites and then experimentally validated the designed helium arrangements. Next, I will present a lightweight-yet-robust machine-learning (ML) pipeline for structure-to-property predictions, where we extract embeddings from a foundation model of neural-network interatomic potentials and use them as descriptors to train shallow ML models for downstream prediction tasks. Together, this talk outlines a streamlined approach to atomic-scale materials design. |