Data-Driven Discovery of Multifunctional Organic Polymer Aerogels by Machine Learning-Assisted Large-Scale Property Screening

Polyimide (PI) aerogels are next-generation thermal insulation materials. Due to their intrinsic nanoporous structures, they are ultralow density and have superior thermal stability, making them important materials in fields such as aeronautics, thermal management, energy storage, and life sciences. The material properties and performance of PI aerogels are strongly correlated to the types of monomers and cross-linking agents. However, the time-consuming and costly synthesis process poses challenges in exploring new formulations of multifunctional PI aerogels with optimized properties. To discover new aerogel formulations, this research utilizes different machine learning methods to model the relationship between pristine monomers, processing conditions, and the properties of cross-linked aerogels. In addition, we have utilized machine learning to discover and predict new chemical structures via large-scale property screening. By introducing the Morgan fingerprint method, the chemical structure, including the hybrid monomer backbone, was effectively used as input for machine learning. Random forest and Gaussian process regression models were trained to estimate the density, compression modulus, and decomposition temperature. Over 1,600,000 novel PI aerogel structures were generated and screened using the trained machine learning models. Consequently, several structural candidates were identified as multifunctional PI aerogels with optimized properties and will have superior performance to those of previously reported aerogels.