Development of a unique multi-layer perceptron neural architecture and mathematical model for predicting thermal conductivity of distilled water based nanofluids using experimental data

Abstract

The thermal conductivity of distilled water based Al2O3, MWCNT, GnP, Al2O3+GnP (50:50) and Al2O3+MWCNT (50:50) nanofluids were experimentally measured. The nanofluid concentration and temperature was ranging from 0.1 to 1 vol% and 30–80 ℃ respectively. Thermal conductivity results from experiments vary non-linearly for nanofluid samples. The enhancement of 17.29%, 24.45%, 22.06%, 18.7% and 20.42% in thermal conductivity was observed using Al2O3, MWCNT, GnP, Al2O3+GnP, Al2O3+MWCNT nanofluids respectively. Based on experimental results, a mathematical model for thermal conductivity has been proposed considering effect of temperature and nanofluid concentration using curve fitting technique. The statistical analysis shows that the mathematical model accurately predicts thermal conductivity of nanofluids for different operating conditions having R2 ranging 0.9945–0.9971. A unique multi-layer perceptron (MLP) artificial neural network (ANN) model having single hidden layer with 21 neurons is proposed. Nanofluids, temperature and concentration form input layer and thermal conductivity as output from ANN. The neural network has least mean square error (MSE) and maximum R2 value of 3.53614e-7 and 0.999 respectively at 21 neurons. Lastly, the experimental data, correlation output and ANN output are compared and found to be in good agreement. The present study can be beneficial to predict thermal conductivity with high accuracy for heat transfer applications.