Estimation of Electrical Conductivity of Molten Multicomponent Slag by Neural Network Computation

Abstract

Information on the electrical conductivity of high-temperature molten oxides can be used as very useful information not only for conventional metallurgical processes on the Earth, but also for the development of extraterrestrial resources such as lunar soil. The models for estimating electrical conductivity of molten slag developed so far apply only to a limited number of slag systems. In this study, a neural network modeling was established to reliably estimate the electrical conductivity of multi-component slag containing TiO₂. In neural network modeling, the electrical conductivity of multi-component slag was used as teaching data to conduct the learning process and Bayesian optimization was applied to determine the estimation precision. As a result, it was possible to construct a reliable neural network estimation model for electrical conductivity that was highly consistent with experimental values collected from the literature data, and the effect of oxide addition on electrical conductivity was investigated using the estimation model. This suggests that it is possible to build a reliable electrical conductivity estimation model by the neural network modeling of accumulated experimental data for multi-component slag, and it is expected that the model will be applied to the electrical conductivity estimation of unknown multi-component slag systems.