Ascertaining Uncertain Nanopore Boundaries in 2D Images of Porous Materials for Accurate 3D Microstructural Reconstruction and Heat Transfer Performance Prediction

Abstract

A precise three-dimensional (3D) microstructure is crucial in porous material-based sciences and technologies, whereas accurate and efficient microstructure reconstruction using two-dimensional (2D) images remains challenging. Here, the strategy of ascertaining nanopore boundaries in focused ion beam-scanning electron microscopy images using experimentally guided image segmentation and deep learning-based reconstruction is proposed for precise reconstruction and heat transfer performance prediction. We demonstrate that uncertain boundaries in 2D images and pore characteristics in reconstructed 3D microstructures can be mathematically linked through area proportion (AP)-determined image segmentation. By calibrating the AP value once using reliable experimental data, accurate reconstruction of porous materials with different pore structures can be achieved with pore characteristic errors of <3%. Compared to randomly generated models, microstructures reconstructed using the proposed method have a better description in the connectivity of solid particles. Porosity-related transport property simulations of reconstructed microstructures reveal the remarkable boundary-identifying ability of the proposed strategy. Benefiting from ascertained porous structures, thermal conductivities in different directions could be predicted using 2D images with accuracy >99% and efficiency <1 s.