Data-Driven Property Prediction of Cellulose Nanofiber/h-BN Composites via RVE Modeling and Machine Learning

Sustainable materials with high mechanical strength and thermal conductivity are essential for next-generation high-density electronics and for reducing electronic waste. Cellulose nanofiber (CNF)-based composites incorporating thermally conductive yet electrically insulating hexagonal boron nitride (h-BN) fillers represent a promising solution to this challenge. In this work, an integrated workflow combining experiments, representative volume element (RVE) modeling, and machine-learning (ML) prediction is developed to design high-performance CNF/h-BN nanopapers. A limited number of experimental datasets are used to construct numerically reliable RVE models. Based on the RVE-derived data, a random forest-based ML framework is established to predict Young's modulus and thermal conductivity for CNF/h-BN composites containing h-BN fillers with different characteristic sizes and porosity. The results reveal that porosity significantly degrades both Young's modulus and thermal conductivity, while the synergistic incorporation of h-BN fillers with different sizes enables concurrent enhancement of mechanical strength and thermal transport. Overall, this study provides a scalable, data-driven strategy for the efficient design and optimization of sustainable, multifunctional composite materials.