MSE Seminar: Applying Data Analytics to Materials Science and Engineering
Speaker: Dongwon Shin, R&D Staff Scientist, Materials Science & Technology Division
Oak Ridge National Laboratory
Title: Applying Data Analytics to Materials Science and Engineering:
What Is (Not) Possible and What Is Needed?
Emerging data analytics techniques certainly have brought high anticipation to significantly accelerate the research and development of many scientific disciplines, including materials science. Contrary to such optimistic expectations, most of the domain scientists, particularly those who do not have a strong background in data science, have questions regarding the deployment of new tools: What can be practically achieved with Al/machine learning, and what are prerequisites? This presentation will attempt to answer these two critical questions in three aspects: 1) preparing datasets constituted with clear input and output, 2) analyzing the correlation between input features and target properties to generate/validate scientific research hypotheses, and 3) performing data analytics. An easy-to-use open-source frontend recently developed by Oak Ridge National Laboratory will be used to analyze the correlation, train machine learning models, and make predictions with trained surrogate models. The examples to be introduced include mechanical and oxidation properties of high-temperature alloys, thermochemical and thermophysical properties of complex oxides, and vapor processing conditions computed from high-throughput computer fluid dynamics simulations performed on a world-class supercomputer.
Dongwon Shin is a Senior R&D Staff Member at Oak Ridge National Laboratory (ORNL), USA. He got his Ph.D. degree from Penn State University in 2007 and spent two years at Northwestern University as a post-doctoral research fellow before joining ORNL as an Alvin M. Weinberg fellow in 2010. His research expertise is CALPHAD and first-principles calculations and recently became interested in applying modem data analytics and supercomputing to design advanced materials. His team has recently developed an open-source frontend, ASCENDS (Advanced data SCiEnce toolkit for Non-Data Scientists), which significantly lowers the barrier of applying emerging data analytics in scientific research, particularly for the scientists and engineers who do not have a strong background in computer/data science. His recent publications are focused on coupling physics to predict the properties of high-temperature alloys within the context of machine learning.