A Hybrid Physics-Based and Data-Driven Approach for Predicting the Effective Thermal Conductivity of Heterogeneous Solids

Abstract

Thermal conductivity is one of the most important physical properties of materials. It plays a significant role in operation, performance and efficiency of the nuclear reactors. This study introduces a novel model for the effective thermal conductivity of polycrystalline solids based on the thin-interface description of grain boundaries (GBs). In contrast to existing models, the new model treats a GB as an autonomous “phase” with its own thermal conductivity. The Kapitza resistance/conductance of a thin interface is then derived in terms of the interface thermal conductivity and width. The predictions of the new model deviate from the corresponding ones from existing models by 1-100% as the grain size approaches the GB width. The development and implementation of two quantitative mesoscale models for the effective thermal conductivity of two important types of nuclear fuels are undertaken. These models account for the effects of temperature, underlying microstructure, and interface thermal resistance for calculating the effective thermal conductivity. High-fidelity finite-element simulations were conducted to validate the predictions of the developed models. These simulations proved the higher accuracy of the developed models. Lastly, to reduce the required computational power, advanced machine learning algorithms were integrated with the validated mesoscale models. This approach is novel and significantly saved the running time and computational cost. The advantages and limitations of the developed models are summarized, and some future directions are highlighted.