Predictability Evaluation of Artificial Neural Networks and Response Surface Methodology Models for Thermo-physical Properties of Graphene Nanoplatelets–Ethylene Glycol/Water Nanofluids for Heat Transfer Applications

Abstract

In this paper, the effectiveness of artificial neural networks (ANN) and response surface methodology (RSM) models in predicting the thermophysical properties ratio of graphene nanoplatelet (GNP)-ethylene glycol (EG)/water nanofluid has been discussed. Volume concentration (0.1%–0.5%) and temperature (−15 °C to 15 °C) were considered as inputs to train the models to predict the thermophysical properties ratios, including density, viscosity, thermal conductivity, and specific heat capacity. The ANN model with the Levenberg–Marquardt (trainlm) algorithm is used to get the best network by varying the number of 9 neurons in the hidden layer. In addition, an RSM, a three-dimensional surface plot techniques technique, was employed on the data points to obtain the new mathematical correlation for predicting thermophysical properties. Eventually, the mean squared error (MSE), regression coefficient (R²), and percentage of errors from both techniques were compared. The proposed ANN and RSM models show that the MSE, R², and percentage of errors are 2.1239 × 10⁻⁵, 0.998, −1.42 to 1.28, and 0.761, above 0.945, −1.46 to 0.97, respectively. The results revealed both techniques are sorely suitable for predicting the thermophysical properties ratio of GNP-EG/water nanofluid.