Machine learning glass transition temperature of polymers

• Published in: Heliyon Vol. 6; no. 10; p. e05055
• Main Authors: Zhang, Yun, Xu, Xiaojie
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01.10.2020; Elsevier
• Subjects: Gaussian process regression; Glass transition temperature; Machine learning; Materials chemistry; Materials science; normal distribution; Physical chemistry; Polymer; regression analysis; Physical chemistry; Machine learning; Gaussian process regression; Materials science; Polymer; Glass transition temperature; Materials chemistry
• ISSN: 2405-8440; 2405-8440
• DOI: 10.1016/j.heliyon.2020.e05055

As an important thermophysical property, polymers' glass transition temperature, Tg, could sometimes be difficult to determine experimentally. Modeling methods, particularly data-driven approaches, are promising alternatives to predictions of Tg in a fast and robust way. The molecular traceless quadrupole moment and molecule average hexadecapole moment are closely correlated with polymers' Tg. In the current work, these two parameters are used as descriptors in the Gaussian process regression model to predict Tg. We investigate 60 samples with Tg values from 194 K to 440 K. The model provides rapid and low-cost Tg estimations with high accuracy and stability. Materials science; Materials chemistry; Physical chemistry; Glass transition temperature; Polymer; Machine learning; Gaussian process regression