Abstract

Machine learning and AI-driven approaches to evaluating materials are highly efficient but can suffer from reduced accuracy and interpretability that is provided by physics-based computational solutions; however, these results may be sufficient to identify material chemistries and microstructures that merit further exploration. Such data-driven approaches are enabled by recent advances in high-throughput experimental techniques that offer exciting opportunities to generate statistically significant quantities of materials characterization data. Similar trade-offs are found in high-throughput experiments, which may miss some of the relevant physics but provide an assessment of whether material performance changes when moving from one specimen to another. By providing a rapid evaluation of new materials, machine learning models support accelerated screening and decision-making for control and optimization of high-throughput processes on the path to materials design. This talk will provide an overview of the AI for Materials Design (AIMD) facility at Johns Hopkins, which highlights some of the challenges, pitfalls and opportunities inherent in an integrated high-throughput and automated materials design framework, in particular addressing challenges associated with assessing high-temperature, high-rate and high-pressure environments. The role of machine learning models in guiding this automated materials design is highlighted and discussed in the context of a few example applications.