Machine Learning-Accelerated Molecular Design of Multi-Functional Polymers—Shifting from Thomas Edison to Iron Man

Polyimides have been widely used in modern industries because of their excellent mechanical and thermal properties, e.g., high-temperature fuel cells, displays, and aerospace composites. Successfully developing a high-performance polyimide usually takes decades of experimental efforts, which is very slow and expensive. However, with the increasing number of experimental data available, materials design and discovery can be accelerated by computational methods like machine learning (ML) techniques and molecular dynamics (MD) simulations in a much more efficient manner. Aiming to discover novel polyimides with multi-functional properties, we first build a comprehensive library of more than 8 million hypothetical polyimides based on the polycondensation of existing dianhydride and diamine/diisocyanate molecules. We then establish multiple ML models for thermal and mechanical properties of polyimides based on their experimentally reported values, including glass transition temperature (Tg), Young’s modulus (<i>E</i>), and tensile yield strength (σy). The obtained ML models demonstrate excellent predictive performance and identify the key chemical substructures influencing the thermal and mechanical properties of polyimides. Applying the well-trained ML models, we obtain predictions of the Tg, <i>E</i>, and σy values of 8 million hypothetical polyimides. In such way, we search the whole hypothetical dataset and identify 3 novel polyimides that have better combinations of Tg, <i>E</i>, and σy than existing ones through Pareto frontier analysis. Furthermore, we validate the ML predictions through all-atom MD simulations and examine their experimental synthesizability. The MD simulations are in good agreement with the ML predictions, and the three novel polyimides are found easy to synthesize. Our study demonstrates an efficient way to expedite the discovery of novel polymers using ML and MD techniques. The exhausting search of a large computational dataset can serve as a general method for other polymeric material exploration problems, such as organic photovoltaics, polymer membranes, and dielectrics.