Liquidus temperature nonlinear modeling of silicates $SiO_2-R_2O-RO$

TL;DR

This paper explores nonlinear modeling of liquidus temperature in silicates using neural networks, demonstrating effective dimensionality reduction and accurate predictions for thermophysical properties.

Contribution

It introduces a nonlinear neural network approach for modeling liquidus temperature, highlighting the use of structural parameters and dimensionality reduction techniques.

Findings

Neural network models outperform linear models in accuracy.

Structural parameters effectively model thermophysical properties.

Dimensionality reduction simplifies complex thermophysical modeling.

Abstract

The liquidus temperature is an important parameter in understanding the crystalline behavior of materials and in the operation of blast furnaces. Its modeling can be carried out by linear and nonlinear methods through data, considering the artificial neural network a modeling method with high efficiency because it presents the theorem of universal approximation and with that better performances and possibility of greater oscillations. The best linear model and the best nonlinear model were modeled by structural parameters and presented a good numerical approximation, thus demonstrating that mathematical modeling can be performed using structural arguments and also showing a dimensionality reduction method for modeling a thermophysical property of the materials.