The thermal conductivity of slag is critical for the efficiency of high temperature metallurgical
processes such as the smelting reduction of ferroalloys, the ladle refining and the continuous casting
mold flux in steelmaking. Experimental determination of slag’s thermal conductivity is fraught with
challenges, as seen in yielding data for the SiO2-Al2O3-CaO system that are often inconsistent and
widely dispersed. In molten slags of this system, thermal transport occurs via phonons within the
network structure. To quantitatively describe thermal conductivity based on the microstructure, hightemperature
Raman spectroscopy was employed to measure the silicate tetrahedra Qi (i = 0, 1, 2,
3), with Q4 species inferred from mass balance. This study also integrated molecular dynamics
simulations to provide a fundamental understanding of the microstructure units distributions.
Machine learning algorithms, especially deep neural networks, were applied to establish the
relationship between slag composition and the configuration of tetrahedra. The thermal
conductivities were then connected with silicate tetrahedra Qi using an Arrhenius-type formalism.
This study sets the stage for extending our findings to more complex, multicomponent slags,
enhancing their practical application in industrial processes.