简 历：Prof. Yi LIU obtained his Ph. D. degree at Materials Science and Engineering at Institute of Metal Research in China in 1997. Then he has worked in the field of computational materials science at Nagoya University, Japan (1997-2002); Juelich Research Center, Germany (2002-2003); University of Western Ontario, Canada (2003-2005); California Institute of Technology, US (2006-2012). He was a professor at the School of Materials Science and Engineering, the University of Shanghai for Science and Technology between 2012-2015 before he moved to Materials Genome Institute and Department of Physics at Shanghai University (2015-present). His current research interests focus on the materials design for superalloy, combustion, nanomaterials, and catalysis by combining computation (density functional theory and molecular dynamics simulations), machine learning, and high-throughput experiment approaches based on materials genome concept.
题 目：Machine Learning Assisted First-Principles Study for Alloying Element Occupation in Superalloy
摘 要：For composition design of multi-component alloy, it is critical to clarifing the preferential site occupancy of alloying elements to elaborate their strengthening mechanisms. It is, however, a formidable task for first-principles (FP) calculations to explore the enormous potential doping configurations in the complex multi-component alloys. In this work, we first carried out high-throughput FP calculations systematically for several hundred alloy doping configurations in superalloy, considering ~10 alloying element substitution at multiple nonequivalent sites. The machine learning models were used for further prediction, reducing the cost of expensive FP calculations while maintaining the certain accuracy. We developed the machine learning models based on the high-throughput FP calculated data. We designed a “Center-Environment” (CE) descriptor model to construct descriptive features by combining elemental properties and local composition and structure information of both center and environment atoms. It is shown that the CE descriptors can be used to predict both the substitution energy and local geometry change of alloying elements in superalloy. By comparison we show clearly that the machine learning prediction using feature construction with both composition and structure information is more accurate and robust than that with composition only. Taking the advantages of the accuracy of first-principle calculations and efficiency of machine learning methods, such combined FP-ML approach becomes an emerging strategy to explore enormous configurations commonly required in computational materials design.