Data Science for Materials in Extreme Environments

"This report examines the value of statistical learning and data mining approaches to the design, development, and modeling of materials in extreme environments, which is a cross-cutting problem within the DoD. Existing tools and databases are briefly surveyed, along with challenges implementing statistical learning methods in the DoD materials research and testing communities. Multi-label classification and microstructural variable quantification are highlighted as important techniques for understanding materials in extreme environments and developing useful processing-structure-property linkages. Case studies exploring classification of high entropy alloy structure, prediction of radiation-induced embrittlement, and computational image processing to characterize material microstructure are presented. The need to document and standardize image analysis methods is emphasized. Barriers to enhancing modeling and design of materials for extreme environments using statistical learning approaches remain, including limited data availability, standardization, and analysis tools, and limited workforce proficiency in techniques for data storage, curation, and analysis."