Unified Multiscale and Explainable Machine Learning Framework for Wear-Regime Transitions in MWCNT and Nanoclay-Reinforced Sustainable Bio-Based Epoxy Composites

This study develops a unified multiscale–machine learning framework to interpret and predict thermo-mechanical wear regime transitions in MWCNT- and nanoclay-reinforced bio-based epoxy composites. A physics-informed master wear formulation integrating real contact mechanics, geometry-dependent shear transfer, interfacial adhesion energetics, and fracture-controlled matrix detachment was combined with interpretable machine learning analytics on a unified tribological dataset. In the CNT system, increasing loading from 0.1 to 0.4 wt.% enhanced interfacial adhesion energy density from 0.00813 to 0.01906 J/m2, resulting in a monotonic reduction in the wear rate from 0.00918 to 0.00613 mm3/N·m (~33% reduction). In contrast, nanoclay exhibited an optimum behavior, with a minimum wear at 0.25 wt.% (0.000093 mm3/N·m; 7.9% reduction vs. neat clay baseline), followed by deterioration at a higher loading due to dispersion loss. The unified probabilistic regime classification of low-wear conditions (k < 0.007 mm3/N·m) achieved an ROC − AUC = 0.9256 and balanced accuracy = 94.3%, with thermo-mechanical severity identified as the dominant regime-switching driver. Reinforcement identity significantly modulated regime stability, confirming distinct shear transfer (Carbon Nano Tubes(CNT)) and confinement/tribofilm (clay) mechanisms within a common mathematical framework. By enabling the durability-oriented design of bio-based tribological systems and extending component service life through predictive stability mapping, this work contributes to resource-efficient materials engineering and reduced lifecycle waste, supporting Sustainable Development Goals SDG 9 (Industry, Innovation and Infrastructure), SDG 12 (Responsible Consumption and Production), and SDG 13 (Climate Action).