Predicting the viscosity of basalt melt by data-driven and interpretable machine learning

Abstract

Basalt fiber is a high-performance fiber made from natural basalt ore by high-temperature melting and filament-forming. The viscosity of basalt melt plays crucial role in regulating melting process and enhancing properties of formed fiber. Here, a dataset of oxide composition in basalt, temperature, and corresponding melt viscosity was collected from reported papers and self-tested samples. By using data-driven and interpretable machine learning technique, two models of Random Forest and Gradient Boosting Decision Tree were established. Both models could learn the dataset and predicted the melt viscosity from the input oxide composition and temperature. A Shapley additive interpretation was conducted on built models, which led to an understanding of significance and pattern of various oxide compositions that impact viscosity. Based on these findings, a prediction on temperature parameters for ore melting and filament-forming was achieved, and continuous basalt fibers were obtained on a fiber spinning facility by using self-tested samples.