Interpretable active learning meta-modeling for the association dynamics of telechelic polymers on colloidal particles

作者：Jalal Abdolahi, Dominic Robe, Ronald G. Larson, Michael Kirley, Elnaz Hajizadeh · 发表于：Journal of Rheology · 年份：2025 · DOI：10.1122/8.0000930 · 被引用次数：7 · 研究领域：Microfluidic and Capillary Electrophoresis Applications、Rheology and Fluid Dynamics Studies

The formulation of waterborne coating fluids composed of colloidal particles interacting with associative polymers as rheology modifiers is a complex multiscale problem with competing design requirements, for which the selection of new formulations remains largely empirical. To move toward rational multiscale design, we here develop active learning meta-models that capture from fine-grained molecular/colloidal simulations the association thermodynamics and dynamics of associative polymers bridging the particles. These properties dictate the macroscopic rheological behavior of these suspensions and can be used in a coarse-grained rheological model for practical predictions. The meta-models were developed using an intelligent search algorithm based on fine-grained data acquired from detailed Brownian dynamics simulations. The active learning approach enables an efficient meta-model development, removing the need for the conventional exhaustive exploration of the entire multidimensional design space. We applied Shapley additive explanations, a machine learning interpretability tool to the developed meta-models, which reveals quantitatively how the gap between particles largely determines the bridge and loop lifetimes. The attraction strength between the polymer ends and particles has minimal effect on the bridge and loop fractions but strongly influences rates of transition between them. These methods pave the way for the computational design of waterborne coatings, guiding benc...