Modelling the effective thermal conductivity and porosity of an open-cell material using an image-based technique coupled with machine learning

Effective thermal conductivity is considered one of the most important parameters in the characterization of insulating materials. The effective thermal conductivity of open-cell porous materials is influenced by both their microstructure and the types of solid and fluid phases present. Currently, many analytical models are available for predicting the final performance based on material porosity, as well as the thermal conductivity of the solid and fluid phases. However, these procedures have two main drawbacks: difficulty in selecting the correct prediction model and challenges in accurately determining porosity. Moreover, when attempting to measure effective thermal conductivity, strict procedural conditions and defined sample geometries are required, necessitating multiple samples due to variations in porosity from specimen to specimen. The procedure proposed here aims to address these challenges. Utilizing microtomography coupled with 3D virtualization and finite element method simulation, this procedure provides validated results which are then used as a dataset to train a machine-learning model capable of predicting thermal conductivity across a comprehensive range of porosities, solid thermal conductivities and fluid thermal conductivities. Comparative analysis has demonstrated that this procedure is significantly more accurate than conventional analytical models or measurement procedures and it has the potential to be applied to any material.