Hierarchical MCTS-Guided Bayesian Optimization for Co-designing Formulation and Processing of Thermosetting Polymers with Balanced Performance

Optimizing the properties of thermosetting polymers presents an inherent challenge due to the high-dimensional formulation-processing parameter space and the trade-offs between thermomechanical properties. In this contribution, we propose a hierarchical Monte Carlo tree search (MCTS)-guided Bayesian optimization (BO) framework to enable the simultaneous design of both the composition and curing protocols of epoxy resins as a major class of thermosetting polymers. A sigmoid-based weighting strategy is introduced to encode application-driven performance thresholds into the objective function, guiding the optimization toward coordinated multi-property improvement. With only 30 experiments across four iterative rounds, the framework optimizes both the formulation and curing protocols to yield a performance-enhanced epoxy resin that simultaneously meets all four target thresholds. This balanced performance cannot be achieved by any of the initial samples, an equal-weight baseline, or commercial and literature-reported systems. Furthermore, interpretable machine learning analysis combined with targeted experimental validation quantitatively resolves the relative importance of formulation and processing parameters, indicating that composition variables dominate overall property trends while their synergistic interactions produce non-linear effects on glass transition temperature. The findings demonstrate that the hierarchical MCTS-guided BO framework offers an efficient strategy for navigating complex formulation-processing spaces, and the interpretable analysis provides a quantitative basis for understanding structure-property relationships in thermosetting polymers.