Interfacial pull-out and failure prediction in MWCNT-functionalized aramid fabric material using machine learning
Makaleler > Tam Makale · 2026
Yazarlar: Havva Esra Bakbak, Erman Bilisik, Kadir Bilisik
Materials Today Communications, Uluslararası, Wos, 2026, cilt 55, ss 115909-115909
DOI: 10.1016/j.mtcomm.2026.115909
Özet
Yarn pull-out is a key energy dissipation mechanism in soft ballistic fabrics; however, its multistage, friction-driven nature challenges predictive modeling. Here, a data-efficient and physics-consistent machine learning framework is proposed to predict the pull-out behavior of MWCNT-modified para-aramid fabrics under single- and multi-yarn extraction. Two complementary approaches, single-stage data-driven learning (SSD-ML) and a physics-guided hybrid machine learning (PG-HML), are comparatively evaluated. The pull-out response is decomposed into crimp extension, initial interlacement rupture, and stick–slip stages, enabling physically interpretable prediction of force–displacement evolution. SSD-ML captures global trends, predicting monotonic increases in maximum pull-out force with increasing MWCNT content (e.g., ~25–112 N for single yarn and ~95–217 N for two yarns). In contrast, PG-HML provides higher fidelity by resolving stage interactions, particularly under multi-yarn extraction, where nonlinear amplification becomes dominant (up to ~1240 N for three yarns). The proposed PG-HML framework achieved high predictive accuracy (R² > 0.98 in Stages I-II), and improved representation of the average Stage III frictional decay response compared with the SSD-ML baseline model. Energy analysis shows that nano-reinforcement significantly enhances frictional resistance. A dimensionless failure criterion is further proposed to relate friction-amplified pull-out force to yarn tensile capacity, enabling prediction of critical yarn-end number and nano-content thresholds governing the transition from stable sliding to tensile rupture. These results suggest that physics-guided machine learning enables physically constrained response prediction and mechanically interpretable representation of complex pull-out behavior, providing a framework for evaluating nano-modified soft ballistic materials.