Machine-learning-assisted Materials Genome Approach for Designing High-performance Thermosetting Polyimides

Enhancing the mechanical properties is crucial for polyimide films, but the mechanical properties (Young’s modulus, tensile strength, and elongation at break) mutually constrain each other, complicating simultaneous enhancement via traditional trial-and-error methods. In this work, we proposed a materials genome approach to design and screen phenylethynyl-terminated polyimides for films with enhanced mechanical properties. We first established machine learning models to predict Young’s modulus, tensile strength, and elongation at break to explore the chemical space containing thousands of candidate structures. The accuracies of the machine learning models were verified by molecular dynamics simulations on screened polyimides and experimental testing on three representative polyimide films. The performance advantages of the best-selected polyimides were analyzed by comparing well-known polyimides based on molecular dynamics simulations, and the structural rationale was revealed by “gene” analysis and feature importance evaluation. This work provides a cost-effective strategy for designing polyimide films with enhanced mechanical properties.