Applying Machine Learning and Quantum Chemistry to Predict the Glass Transition Temperatures of Polymers

Developing models to predict glass transition temperatures (Tgs) of polymers is of significant importance given how the parameter quantifies the physical and thermal characteristics of these materials. These characteristics inform numerous functional properties of polymers as well as how they degrade both through intended use as well as through environmental mechanisms that result from end-of-life environmental deposition. For this reason, various models have been developed for predicting Tg from structural information to aid in designing novel polymer materials. These existing models, however, typically focus on utilizing one specific modeling technique to train a single class or set of polymers. To expand and explore the applicability of Tg models for predictions of new materials this work (1) utilizes both machine learning (ML) and quantum chemistry (QC) based techniques to investigate different data availability scenarios and (2) does not constrain the training datasets to any specific class of polymers. This methodology allows for a comparison of different techniques and situations to determine the applicability of Tg models when making predictions for novel polymer structures.

开发用于预测聚合物玻璃化转变温度（glass transition temperatures, Tgs）的模型具有重要意义，该参数可量化这类材料的物理与热学特性。这些特性不仅决定了聚合物的诸多功能属性，还决定了其在实际使用过程以及终端生命周期环境沉积所引发的环境机制下的降解行为。鉴于此，学界已开发出多种基于结构信息预测Tgs的模型，以助力新型聚合物材料的设计研发。然而，现有模型通常仅依托单一特定建模技术，针对单一类别的聚合物或某一组聚合物开展训练。为拓展并探究Tgs模型在新型材料预测中的适用性，本研究（1）同时采用基于机器学习（machine learning, ML）与量子化学（quantum chemistry, QC）的技术，针对不同数据可得性场景开展研究；（2）未将训练数据集限定于任何特定类别的聚合物。该研究方法可实现不同技术与应用场景间的对比，以明确Tgs模型在预测新型聚合物结构时的适用范围。