Efficient exploration of compositional space for high-performance copolymers via Bayesian optimization

Abstract

The traditional approach employed in copolymer compositional design, which relies on trial-and-error, faces low-efficiency and high-cost obstacles when attempting to simultaneously improve multiple conflicting properties. For example, designing co-cured polycyanurates that exhibit both moisture and thermal resistance, along with high modulus, is a long-term challenge because of the intrinsic trade-offs between these properties. In this work, to surmount these barriers, we developed a Bayesian optimization (BO)-guided method to expedite the discovery of co-cured polycyanurates exhibiting low water uptake, coupled with higher glass transition temperature and Young's modulus. By virtue of the knowledge of molecular simulations, benchmarking studies were carried out to develop an effective BO-guided method. Propelled by the developed method, several copolymers with improved comprehensive properties were obtained experimentally in a few iterations. This work provides guidance for efficiently designing other high-performance copolymers.