RUS  ENG
Full version
JOURNALS // Computer Research and Modeling // Archive

Computer Research and Modeling, 2024 Volume 16, Issue 5, Pages 1217–1252 (Mi crm1215)

MODELS IN PHYSICS AND TECHNOLOGY

Modeling of rheological characteristics of aqueous suspensions based on nanoscale silicon dioxide particles

L. N. Marchenkoab, Ya. A. Kosenoka, V. E. Gaishuna, Yu. V. Bruttancb

a Francisk Skorina Gomel State University, 104 Sovetskaya st., Gomel, 246028, Belarus
b Scientific and Educational Mathematical Center “Sofia Kovalevskaya Northwestern Center for Mathematical Research” in Pskov State University, 2 Lenin sq., Pskov, 180000, Russia
c Pskov State University, 2 Lenin sq., Pskov, 180000, Russia

Abstract: The rheological behavior of aqueous suspensions based on nanoscale silicon dioxide particles strongly depends on the dynamic viscosity, which affects directly the use of nanofluids. The purpose of this work is to develop and validate models for predicting dynamic viscosity from independent input parameters: silicon dioxide concentration SiO$_{2}$, pH acidity, and shear rate $\gamma$. The influence of the suspension composition on its dynamic viscosity is analyzed. Groups of suspensions with statistically homogeneous composition have been identified, within which the interchangeability of compositions is possible. It is shown that at low shear rates, the rheological properties of suspensions differ significantly from those obtained at higher speeds. Significant positive correlations of the dynamic viscosity of the suspension with SiO$_{2}$ concentration and pH acidity were established, and negative correlations with the shear rate $\gamma$. Regression models with regularization of the dependence of the dynamic viscosity $\eta$ on the concentrations of SiO$_{2}$, NaOH, H$_{3}$PO$_{4}$, surfactant (surfactant), EDA (ethylenediamine), shear rate $\gamma$ were constructed. For more accurate prediction of dynamic viscosity, the models using algorithms of neural network technologies and machine learning (MLP multilayer perceptron, RBF radial basis function network, SVM support vector method, RF random forest method) were trained. The effectiveness of the constructed models was evaluated using various statistical metrics, including the average absolute approximation error (MAE), the average quadratic error (MSE), the coefficient of determination $R^{2}$, and the average percentage of absolute relative deviation (AARD%). The RF model proved to be the best model in the training and test samples. The contribution of each component to the constructed model is determined. It is shown that the concentration of SiO$_{2}$ has the greatest influence on the dynamic viscosity, followed by pH acidity and shear rate $\gamma$. The accuracy of the proposed models is compared to the accuracy of models previously published. The results confirm that the developed models can be considered as a practical tool for studying the behavior of nanofluids, which use aqueous suspensions based on nanoscale particles of silicon dioxide.

Keywords: nanofluid, SiO$_{2}$ concentration, pH acidity, dynamic viscosity, regression, neural networks, machine learning

UDC: 004.85, 546.28

Received: 27.07.2024
Revised: 16.09.2024
Accepted: 24.09.2024

DOI: 10.20537/2076-7633-2024-16-5-1217-1252



© Steklov Math. Inst. of RAS, 2025