3D X-Ray Tomography-Driven Descriptor Framework for Predicting Thermal Conductivity of Polymer Composites

Polymer composites filled with ceramic particles are widely studied as thermal interface materials (TIMs), yet thermal conductivity gains do not scale linearly with filler loading. Experimental inconsistencies suggest hidden structural factors governing heat transport. Here, 3D X-ray tomography is combined with quantitative morphometry to identify descriptors that control conduction in alumina-polydimethylsiloxane (PDMS) composites. Five composites with varying filler loadings (60-82.5 vol%) and four alumina size fractions (90, 20, 3, and 0.6 µm)-including one Bayesian-optimized formulation-are reconstructed to visualize particle networks and polymer domains. Local thickness analysis revealed how thin polymer ligaments enable continuous pathways, while tortuosity (τ) mapping showed how clustered alumina domains disrupt transport even at high loadings. Surface-to-volume ratio (S/V) further quantified interfacial density, revealing that excessive interface area can hinder conduction. A predictive model integrating filler fraction (ϕ), normalized S/V, and τ identified an optimal 80.3 vol% formulation outperforming the maximum-loading case (82.5 vol%) due to moderated tortuosity and controlled interfacial density. This tomography-derived framework advances from qualitative visualization to quantitative design rules, establishing local thickness as a bridge between geometry and transport, and offering a transferable pathway for rational microstructure engineering in TIMs and multifunctional composites.