TITLE:
Modeling and Prediction of the Melting Temperature of Soda-Lime Glasses Based on Their Chemical Composition Using Artificial Neural Networks
AUTHORS:
Mohamed Karamoko, Péyokoh Roger Thio, Kouassi Bruno Koffi, Kouadio Denis Konan
KEYWORDS:
Soda-Lime Glass, Melting Temperature, Artificial Neural Network, Machine Learning, SiO2-Na2O-CaO Ternary System
JOURNAL NAME:
Journal of Materials Science and Chemical Engineering,
Vol.14 No.8,
August
25,
2026
ABSTRACT: The melting temperature of soda-lime glasses is a key processing parameter that governs energy consumption, melt viscosity, and manufacturing cost. Owing to the highly nonlinear relationship between glass composition and melting behavior, accurate prediction of melting temperature remains challenging. In this study, an artificial neural network (ANN) was developed to predict the melting temperature of glasses in the SiO2-Na2O-CaO ternary system directly from their chemical composition. A database containing 45 unique compositions collected from the literature and experimental studies was used for model development. The ANN employed the molar fractions of SiO2, Na2O, and CaO as input variables and the melting temperature as the output. The proposed 3-10-10-1 feed-forward architecture achieved a mean absolute error (MAE) of 2.74˚C and a root mean square error (RMSE) of 4.66˚C on the training dataset. External validation yielded an MAE of 18.29˚C and an RMSE of 20.68˚C, while an additional validation using independent experimental data reported by Santoso et al. confirmed the predictive capability of the model, with an MAE of 14.35˚C and an RMSE of 19.12˚C. These results demonstrate that the proposed ANN accurately captures the relationship between chemical composition and melting temperature and constitutes a reliable and efficient tool for the preliminary design and optimization of soda-lime glass compositions, thereby reducing the need for extensive experimental investigations.