Network analysis of heat transfer in sphere packings

Abstract

Porosity-dependent models can be used to predict the effective thermal conductivity (ETC) of particulate materials. However, they cannot directly account for microstructural features such as particle connectivity and interparticle contact area. Complex network theory can be used to extract network features as microstructural characteristics. However, these features have not been used to study heat transfer. In this work, both contact network and thermal networks are constructed for mono-disperse and poly-disperse sphere packings. Network features are extracted using complex network theory and machine learning techniques are applied to investigate the correlation between these features and the ETC. The most relevant thermal and contact network features for predicting thermal conductivity are identified. The network features capturing both interparticle connectivity and contact quality, such as “weighted degree”, show high correlation with ETC. Furthermore, random forest regression results show that involving multi-network features in a model enhances the accuracy in predicting ETC.