Artificial neural network-based prediction of effective thermal conductivity of a granular bed in a gaseous environment

Abstract

Artificial neural network (ANN), a machine learning technique, is employed to predict the effective thermal conductivity of granular assemblies in the presence of a stagnant gas. ANN is trained with the help of estimated thermal conductivities calculated through resistor network (RN) model. RN model considers the effect of the presence of stagnant gas and the gas pressure (Smoluchowski effect) for the calculation of effective thermal conductivity. Granular assemblies are generated and compacted through discrete element method (DEM). The ANN is trained to predict the effective thermal conductivity of a granular assembly for a set of measurable experimental parameters (stress and packing fraction) without requiring the knowledge of microstructural details (coordination numbers and overlaps) of the assembly. The predicted effective thermal conductivity values through ANN are in good agreement with the experimental results. Estimation of effective thermal conductivity through the trained ANN is much faster (few seconds compared to few hours required for DEM together with RN approach) with very good accuracy.