Abstract
Abstract
The use of machine learning (ML) models to screen new materials is becoming increasingly common as they accelerate material discovery and increase sustainability. In this work, the chemical structures of 16 epoxy resins and 19 curing agents were used to build an ML ensemble model to predict the glass transition ($$T_g$$
T
g
) of 94 experimentally known thermosets. More than 1400 molecular descriptors were calculated for each molecule, of which 119 were chosen based on feature selection performed by principal component analysis. The quality of the trained model was evaluated using leave-one-out cross-validation, which yielded a mean absolute error of 16.15$$^{\circ }$$
∘
C and an $$R^2$$
R
2
value of 0.86. The trained model was also used to predict $$T_g$$
T
g
for 4 randomly selected resin/hardener combinations for which no experimental data were available. The same combinations were then prepared and measured in the laboratory to further validate the ML model. Excellent agreement was found between experimental and predicted $$T_g$$
T
g
values. The current ML model was created using only theoretical features, but could be further improved by adding experimental or quantum mechanical properties of the individual molecules as well as experimental processing parameters. The results presented here contribute to improving sustainability and accelerating the discovery of novel materials with desired target properties.
Graphical Abstract
Publisher
Springer Science and Business Media LLC
Subject
Mechanical Engineering,Mechanics of Materials,General Materials Science
Cited by
11 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献