Thermal conductivity prediction of Al[formula omitted]O[formula omitted]-doped tetragonal YSZ coatings using deep learning

Abstract

Thermal barrier coatings, including Al2O3-doped tetragonal yttria-stabilized zirconia (t-YSZ) coatings are vital in diverse applications. Thermal conductivity is a key property, but prediction often requires some complex experiments. In this study, a novel dual-structure feature extraction coupled with a multi-scale attention fusion network (RCFNet) is introduced for accurate thermal conductivity prediction using electron backscattered diffraction (EBSD) microstructure images. The method requires minimal data, and the result shows high accuracy with a Mean Square Error (MSE) of 0.003 W(m⋅k)−1 and an R-squared (R2) value of 0.81. Furthermore, the model reveals specific correlations between microstructural characteristics and thermal conductivity, facilitating rapid structure selection for specific thermal conductivity.