Discovery of Trade-Off Performance Frontiers of Cross-Linking Polymers through Artificial Intelligence-Driven Multiobjective Bayesian Optimization

The intrinsic trade-offs among various properties have always hindered the pursuit of materials with excellent overall performance. Traditional trial-and-error methods have revealed inherent limitations in addressing material optimization problems under high-dimensional parameter spaces and complex property interdependencies. Especially, it is a longstanding challenge to simultaneously improve Young’s modulus, strength, and toughness of cross-linking epoxy resins as structural matrices in high-performance materials. In this contribution, by leveraging the multiobjective Bayesian optimization (MOBO) approach, we overcome the challenge through an AI-driven investigation of the optimization of curing conditions, a critical yet often overlooked factor in epoxy resins, to simultaneously achieve multiple outstanding properties. In particular, as a multiobjective-oriented acquisition function, the expected hypervolume improvement is applied to evaluate the properties of epoxy resins under unexplored regions of curing conditions. It is demonstrated that the proposed method can effectively optimize the multivariate curing conditions of epoxy, including comprehensive improvement of all properties or enhancement of a specific property. Importantly, the feature analysis of data-driven surrogate models highlights the critical role of curing conditions at distinct stages in determining the ultimate performance of epoxy resins, originating from a change in microscopic structures. The multiobjective Bayesian optimization method proposed in this work provides a promising AI-based paradigm for accelerating the high-performance materials design in high-dimensional parameter space and understanding the processing–structure–property relationships from data-informed perspectives.