Machine Learning based Software Package for Reproducible Data Analysis for Electron Microscopy

Advances in hardware and computation technologies are allowing electron microscopy (EM) datasets to be produced at rapidly increasing spatial and temporal resolutions, which presents exciting opportunities for microstructure analysis but also great challenges in data analysis. In order to keep pace with the growing tide of data, new analysis tools must be created to enable the analysis of large volume data for users. Under this STTR program, QuesTek Innovations LLC, a leader in Integrated Computational Materials Engineering (ICME), proposes to combine Machine Learning (ML) techniques with ICME tools to develop a software to enable reproducible EM data analysis for multiple EM systems and data types. QuesTek will partner with Dr. Xiaoting Zhong and her team, with extensive research experiences in ML-based microstructure analysis at Lawrence Livermore National Laboratory (LLNL) to develop ML and computer vision (CV) based data analysis models. QuesTek will also collaborate with Dr. Nestor Zaluzec, a lead researcher in electron microscopy at the Center for Nanoscale Materials at Argonne National Laboratory (ANL, a Nanoscale Science Research Center) to use upcoming new PicoProbe equipment for the hyperspectral imaging of hard and soft matter for multimodal and multidimensional operations. Under the Phase I program, a software prototype will be developed focusing on TEM techniques tailored for metallic materials characterization. The key objective of the Phase I program is to generate a software prototype to serve as a proof-of-concept for demonstrating feasibility of reaching the overall program goal. The capabilities of this software prototype will include: ML models for data analysis and phase/structure identification for various data types such as image, diffraction, and spectroscopy, and models to connect microscopically observable features to macroscopic properties. The models will be based on QuesTek’s and LLNL’s in-house computer vision (CV) and ML codes. These ML models will be physics-informed and science-aware by incorporating QuesTek’s proprietary ICME technologies centered on CALculations of Phase Diagram (CALPHAD) which accurately depict thermodynamics and kinetics of the materials systems. A live feedback framework to enable automated experimentation for in-situ experiments (e.g. in- situ precipitation experiment). A cloud-based platform with graphical user interface (GUI) to provide data visualization as well as to enable the connection between the user’s manual input and the backend models. In the Phase II program, we will work closely with Dr. Zaluzec to apply this software package for interactive experiments for a wider range of materials. The software prototype will be validated with a specific use case, leveraging QuesTek’s previous materials design experience and data. In addition, the developed software prototype will be deployed on QuesTek’s Integrated Design Engine (QT-IDE), which is a cloud- based platform with a variety of toolkits for next-generation materials design to enable cloud-based CALPHAD calculations. QuesTek will also expand the capabilities of the software package to other EM systems such as scanning electron microscopy (SEM) and electron backscatter diffraction (EBSD).