Using artificial neural networks to predict the rheological behavior of non-Newtonian graphene–ethylene glycol nanofluid

The ability of the artificial neural network (ANN) to predict the viscosity of graphene nanosheet/ethylene glycol (μGr/EG\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu_{{\text{Gr/EG}}}$$\end{document}) was examined. The nanofluid conformed to the non-Newtonian classification which consequently three neurons were assigned to the temperature, mass fraction and shear rate. Considering the maximum R-squared (R2) as well as the minimum mean square error (MSE), the approved ANN consisting of 10 neurons in the middle layer, had an acceptable performance so that the statistical calculations affirmed that the values of MSE, R2 were 0.97185 and 0.9978, respectively. Although the highest margin of deviation (MOD) was reported to be 6.69%, more than 60% of the input points had the MOD less than 1%. The ability of the ANN to estimate μGr/EG\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu_{{\text{Gr/EG}}}$$\end{document} depends on temperature and mass fractionation, so that as the temperature rises, the amount of MOD increases, which means that at higher temperatures, the accuracy diminishes slightly.