高性能聚酰亚胺材料数据驱动-QSPR预测模型构建

聚酰亚胺材料（Polyimide, PI）由于特殊的主链结构及链间相互作用，表现出优异的热学、力学、电学和耐溶剂性能，已经成为许多高技术产业领域的核心关键材料。目前，聚酰亚胺新材料研制主要采用实验试错模式，工作量巨大、研究周期很长，严重制约着高新技术产业的快速发展。近年来，“材料信息学+机器学习”已经成为高科技材料研制的新范式，可对新结构材料进行性能预判，大幅度缩短新材料的研制周期，对材料领域具有革命性意义。本项目以聚酰亚胺材料已经积累的实测数据为基础，依据单体分子结构及齐聚物模型，构建聚酰亚胺材料的分子结构参数(描述符)数据库，采用机器学习方法建立聚酰亚胺结构与性能定量构效关系(Quantitative Structure-Property Relationship，QSPR)模型，实现对特种聚酰亚胺材料性能的准确预测、新材料的设计及实验指导。

Owing to the special molecular structures of polymer backbones and inter-chain interactions, polyimides (PIs) have many desirable characteristics such as excellent thermal, mechanical, dielectric and optoelectronic properties, and have been extensively applied in many high-tech areas, including electric insulating, microelectronics, optoelectronics, aerospace and etc. The current design approaches for new PI materials, however, mainly based on experimentally-driven, “trial and error” process. It is facing significant challenges due to the tremendous demand of research efforts and long research period. Nowadays, the strategy of “material informatics + machine learning” has become a new paradigm of material research. This strategy considerably reduces the time span on material discovery and shortens the development period, thus has been considered to have revolutionary impacts on the design of key materials. In this project, a data set of polyimides with known experimental values will be first collected. Then, a fingerprinting scheme (descriptors) that captures structural features will be constructed to represent the PI materials. Machine learning models driven by the descriptors will be then developed to create Quantitative Structure-Property Relationship (QSPR) for PI materials. The data-driven QSPR models developed, which can be employed to predict the special properties (such as thermal expansion, optical transparency, high frequency transmission etc.) of PI materials, would be implemented in the discovery process of new materials.