Identifying key parameters for predicting materials with low defect generation efficiency by machine learning

• Published in: Computational materials science 2021-04, Vol.191, p.110306, Article 110306
• Main Authors: Ni, Dongyuan, Wu, Wei, Guo, Yaguang, Gong, Sheng, Wang, Qian
• Format: Article
• Language: English
• Subjects: Low defect generation efficiency; Machine learning
• ISSN: 0927-0256
• DOI: 10.1016/j.commatsci.2021.110306

[Display omitted] •Applying molecular dynamic simulations for studying the defect generation of materials during the primary damage process.•Identifying the key features of primary damage.•Using the well-trained neural network to predict new materials with low defect generation efficiency. The primary radiation damage is an important part of the radiation process, which is of current interest as the rapid development of nuclear reactors and space instrumentation. In this study, using machine learning, we have demonstrated that atomic mass difference, Poisson’s ratio, mean atomic mass, and mass density have significant influence on the defect generation efficiency of a material during the primary damage step. Furthermore, we construct a new dataset by using these important features and obtain a well-trained neural network for predicting new materials with low efficiency of defect generation. In our study, the target of the dataset for training the predictor is constructed using the results from molecular dynamics simulations. This work provides the guiding information for designing materials with low efficiency of defect generation.