Investigation of different training function efficiency in modeling thermal conductivity of TiO2/Water nanofluid using artificial neural network

Abstract

In the present study, using an artificial neural network (ANN), the thermal conductivity of nanofluid containing TiO2 nanoparticles was modeled and investigated. The used experimental data to train and estimate the ANN are the thermal conductivity of the nanofluid at different volume fractions and temperatures. The ANN structure is considered with two hidden layers and five neurons for each hidden layer. To choose the best training function, several training functions available in the ANN are analyzed. The results reveal that by considering the criteria of the highest regression coefficient and the lowest mean square error (MSE), trainbr training function has the best performance with R= 99.9% 2%, MSE= 1.0211e-5 and train and test performances of 3.1198e-6 and 5.2791e-5. The maximum error of 2.5% for test data approximation in the process of knfestimation shows the precision of designed ANN. The results also show that ANN can estimate the knfof TiO2/water nanofluid (NF) at different temperatures and φpwith very good precision, and theoretical models like Pak and Cho are not able to precisely estimate the knf.