Title

VISCOSITY PREDICTION OF SILICA REFRACTORIES USING ARTIFICIAL NEURAL NETWORKS

VISCOSITY PREDICTION OF SILICA REFRACTORIES USING ARTIFICIAL NEURAL NETWORKS

Authorship

Anjos, Patrick Queiroz Dos; Machado, Marcelo Lucas Pereira; Quaresma, Lucas de Almeida

Anjos, Patrick Queiroz Dos

I am this author

Machado, Marcelo Lucas Pereira

I am this author

Quaresma, Lucas de Almeida

I am this author

DOI

10.5151/2594-357X-39580

Downloads

Abstract

The production of silica refractories using fused silica has an advantage over refractories produced by the conventional method, as it does not present volumetric variations due to changes in crystalline phases and because they present the possibility of complex formats. As such, viscosity is critical to the process. Viscosity data were taken from the Sciglass database with chemical system SiO 2 -Al 2 O 3 -CaO- K 2 O-Na 2 O-TiO 2 , temperature and each related viscosity. Numerical modeling was performed using artificial neural networks by varying neurons and hidden layers. The best artificial neural network had 3 neurons and 1 hidden layer and demonstrated collinearity with the test data and the predicted data using the test data. Sensitivity analysis was performed and demonstrated agreement with the literature. The main parameters of the 3-1 artificial neural network for the prediction of viscosity in silica refractories were demonstrated.

The production of silica refractories using fused silica has an advantage over refractories produced by the conventional method, as it does not present volumetric variations due to .changes in crystalline phases and because they present the possibility of complex formats. As such, viscosity is critical to the process. Viscosity data were taken from the Sciglass database with chemical system SiO2-Al2O3-CaO-K2O-Na2O-TiO2, temperature and each related viscosity. Numerical modeling was performed using artificial neural networks by varying neurons and hidden layers. The best artificial neural network had 3 neurons and 1 hidden layer and demonstrated collinearity with the test data and the predicted data using the test data. Sensitivity analysis was performed and demonstrated agreement with the literature. The main parameters of the 3-1 artificial neural network for the prediction of viscosity in silica refractories were demonstrated.

Keywords

Silica refractories; Database; Mathematical modeling; Artificial neural networks.

Silica refractories; Database; Mathematical modeling; Artificial neural networks