文献类型：期刊文献

英文题名：Machine learning approaches for designing polybenzoxazines with balanced thermal stability and dielectric properties

作者：Zhang, Jiahang[1];Yu, Yong[1];Zhuang, Qixin[1];Yin, Wei[2];Zuo, Peiyuan[1];Liu, Xiaoyun[1]

机构：[1]East China Univ Sci & Technol, Lab Specially Funct Polymer Mat & Related Technol, Minist Educ, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Sch Mech & Power Engn, Shanghai 200237, Peoples R China

年份：2025

卷号：68

期号：8

起止页码：3732

外文期刊名：SCIENCE CHINA-CHEMISTRY

收录：;EI(收录号：20251218084008);WOS:【SCI-EXPANDED(收录号:WOS:001448774400001)】；

基金：This work was supported by the National Natural Science Foundation of China (22171086, 52373073, 52073091).

语种：英文

外文关键词：polybenzoxazine; machine learning; thermal stability; dielectric properties

摘要：Polybenzoxazines are widely used as high-performance polymers in machinery, aerospace, and other industries. However, despite recent advances in synthesizing improved polybenzoxazines, achieving a good balance between multiple properties still presents a significant challenge. More specifically, this difficulty arises from the sparsity of historical experimental data and the lack of a well-established structure-property relationship, which hinders the development of polybenzoxazines with excellent overall performance. This study proposes a machine-learning-assisted approach that rapidly screens novel benzoxazines with high thermal stability and excellent dielectric properties by exploring a vast chemical space. Three highly reliable machine learning models are developed to predict the 5% weight loss temperature (Td5), dielectric constant, and dielectric loss of polybenzoxazines, respectively. Subsequently, high-throughput benzoxazines are designed using a reaction template, and property prediction is performed using a machine learning model we created. Then, experiments were carried out to verify the designed structures. The results indicate that the experimental values of the polybenzoxazines align closely with the predicted values from the machine learning model, with errors falling within acceptable limits. In addition, substructures that affect the thermal stability and dielectric properties are also extracted and discussed. Compared to the traditional trial-and-error approach, this new method offers a more efficient and cost-effective way to accelerate the innovation of high-performance thermosetting resins.