Employing ANN Model for Prediction of Thermal Conductivity of CNT nanofluids

Artificial neural network techniques are widely used for prediction purposes. The main advantage of ANN is on the solution of complex non-linear problems. The thermophysical properties of nanofluids depend on many parameters i.e. nature of the base fluid, particle, temperature, volume concentration, shape and size of nanoparticles etc. This dependency makes the relation between thermophysical property and variable parameters very complex and is very difficult for prediction purposes. The accurate prediction of thermophysical properties cannot be done by conventional models, this time artificial neural network technique is suited for accurate prediction. In this work, different % Vol. concentration (0.1-2%) CNT based nanofluids were prepared at various base fluids (transformer oil, Ethylene Glycol (EG), and water). Their thermal conductivity was measured by transient hot-wire setup at different temperatures (20-50 °C). The thermal conductivity results from the measurement were higher than those predicted by the conventional models. Therefore, a Multilayer perceptron (ANN) model has been utilized for accurate prediction of thermal conductivity at different base fluid, concentration and temperature. These variable parameters were selected as input for the ANN model. For the prediction of thermal conductivity in the broad range of parameters. Levenberg -Marquardt algorithm was employed in the present ANN model. The final optimal model can accurately predict the thermal conductivity of CNT nanofluids. The predicted thermal conductivity of the CNT nanofluid of the ANN model coincides with the experimental data. Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and correlation coefficient (R
²
) were the selection criteria of the ANN model. For the final optimal model, their values were 0.005, 0.0033 and 0.9999 respectively. The deviation of results was within ±5%. The present ANN model have a higher prediction capability over conventional models.