Abstract:

Predictive machine learning (ML) tools can accelerate the discovery of new materials and enable semi or fully autonomous research. In addition, ML tools are playing an increasingly central role in physics-based simulations, enabling a combination of accuracy and computational efficiency not possible hitherto. I will present recent progress by my group in these areas and discuss current challenges and opportunities.

Multiscale models for materials at extreme conditions. Materials respond to dynamical or shock loading via a plethora of mechanisms, some of which do not have counterparts under quasi-static or weak loadings. Large-scale molecular dynamics (MD) simulations can describe these processes with few approximations, I will describe our recent work on the response of energetic materials and high-entropy alloys to shock loading. Connecting these detailed atomic simulations to continuum models capable of reaching the scales of interest in most applications has been a longstanding challenge, I will show how tools from ML can be used to address this challenge.

Materials discovery.  I will also describe how ML can accelerate the discovery of new materials. We used active learning, combined with physics-based modeling and fabrication and characterization experiments, to discover alloys for high-temperature applications. This approach resulted in the hardest Al-containing BCC-based alloy with only 24 tests within a design space consisting of ~67,000 possible materials.

FAIR data and workflows for materials discovery. Key to the unleashing of the full power of ML in materials science is data availability. I will describe recent developments in nanoHUB that seek to make simulation workflows and their data findable, accessible, interoperable, and reusable (FAIR). I will introduce Sim2Ls (pronounced sim tools) that allow developers to create and publish end-to-end computational workflows with well-defined and verified inputs and outputs. The workflows are discoverable and available online. Importantly, all inputs and outputs from each execution are automatically stored in a FAIR database. We believe this open infrastructure can accelerate innovation in research and be used to introduce students to modern data practices.

Biography:

Alejandro Strachan is the Reilly Professor of Materials Engineering at Purdue University and the Co-Director of nanoHUB and chipshub. Before joining Purdue, he was a Staff Member in the Theoretical Division of Los Alamos National Laboratory and worked as a Postdoctoral Scholar and Scientist at Caltech. He received a Ph.D. in Physics from the University of Buenos Aires, Argentina.

Prof. Strachan’s research focuses on predictive atomistic and multiscale models to describe materials from first principles and their combination with machine learning and artificial intelligence to address problems of technological or scientific importance. In addition, Strachan’s scholarly work includes cyberinfrastructure to make simulations, models, and data widely accessible and useful for research and education.

Prof. Strachan has published over 200 peer-reviewed scientific papers and his contributions to research and education have been recognized by several awards, including the Early Career Faculty Fellow Award from TMS in 2009, an R&D 100 award in the category of software and services for nanoHUB (2020), and the Reilly Chair Professorship in 2023.