Le Pr. Mithra Taheri, du Department of Materials Science & Engineering de l’Université Johns Hopkins donnera un séminaire au département M2I intitulé :

Atomic Legos: From atoms to machines and back again

le lundi 20 juin 2022 à à 11h00 à l’auditorium.

Abstract :

As the world moves toward autonomous experimentation, microscopy provides a foundational tool on which to develop and build the necessary data science foundation for these toolsets. Tracking dynamic processes using electron microscopy tools- imaging, spectroscopy, diffraction- generates large data sets that are often only analyzed after the experiment is over. In this talk, a path forward toward real-time data processing, in operando, is discussed. Specifically, machine learning approaches for denoising, classification, and related accuracy are discussed in the context of accuracy and decision processes “on the fly.” Examples of operando control of oxides and 2-dimensional MXenes are presented, revealing the ability to tailor operando experiments with precision.

The results presented provide a look into the future of “bespoke” materials and the leading role of next generation electron microscopy in these efforts. Specifically, I will highlight our work exploring local structure evolution in high entropy alloys and metallic glasses, with profound implications in developing thermally stable, damage tolerant materials, with implications in critical applications in energy and aerospace. In this project, we leverage both high throughput and precision techniques to benchmark local and global fingerprints within these high entropy alloy systems that give rise to favorable outcomes in corrosion and oxidation environments, and most importantly, provide a foundation for new alloy discovery and development.

Finally, an outlook on emerging time resolved studies and key challenges for “big data” will be presented. A major challenge in understanding complex phase transformations is obtaining high resolution spectra, images, and other critical property-classifying materials data at rapid time scales. This work will be discussed as a foundation for future integration of ML at high speeds during synthesis and processing experiments, ranging from 3D printing to in situ microscopy.