High-Performance Polyimide Design: Machine Learning-Accelerated Multiobjective Optimization

Polyimides (PIs) with superior transparency, mechanical robustness, and thermal stability are essential for advanced applications such as flexible electronics, foldable displays, and aerospace components, but simultaneous optimization of these properties remains challenging. Here, we present a data-driven multiobjective optimization framework for systematic polyimide design, supported by high-quality data sets extracted with the assistance of large language models. Leveraging robust machine learning models validated by the experimental synthesis and evaluation of five PI films, our method precisely predicts material performance and effectively deciphers complex structure-property relationships influenced by charge transfer complexes. The demonstrated capability of rapidly screening over 20,000 potential PI formulations significantly accelerates material innovation. Delivers PI-a, PI-b, and PI-e films that exceed current commercial benchmarks, achieving glass transition temperature values of 336-376 °C, tensile strengths of 207-324 MPa, and transmittance above 89.6%. This framework establishes a broadly applicable, data-driven approach for designing multifunctional polymers tailored for next-generation flexible electronics and energy storage devices.