Predicting Interfacial Pull-Out Performance of Nano-B4C/Aramid Material with Stage-Wise Physics-Guided Machine Learning

Interfacial yarn pull-out plays a critical role in load transfer and energy dissipation in soft ballistic materials; however, its multistage and friction-dominated nature makes comprehensive experimental characterization time-consuming and experimentally demanding. In this study, a stage-wise physics-guided machine learning (PG-HML) framework is proposed to predict the pull-out behavior of nano hexagonal boron carbide (nh-B4C)-functionalized para-aramid fabrics using an experimentally constrained dataset. The pull-out response was decomposed into three physically meaningful deformation stages, namely crimp extension, initial interlacement rupture, and stick–slip sliding. Physics-based continuity and admissibility constraints were incorporated into the learning framework to preserve mechanical consistency across stage transitions and improve prediction robustness under limited data conditions. Comparative analyses demonstrated that the proposed PG-HML framework achieved superior predictive capability, particularly in capturing rupture transitions and post-peak stick–slip evolution, with R2 values exceeding 0.98 during the crimp extension and rupture stages. Increasing nh-B4C content enhanced interfacial friction, rupture resistance, and pull-out energy dissipation, while displacement responses gradually approached saturation under force-dominated extraction conditions. Therefore, interfacial pull-out behavior in nh-B4C/aramid materials can be predicted with high fidelity using limited experimental input, providing a surrogate modeling strategy that enables virtual material screening and significantly reduces the experimental effort required for preliminary design and optimization of advanced soft ballistic materials.