On the evaluation of thermal conductivity of nanofluids using advanced intelligent models

Accurate knowledge of thermal conductivity (TC) of nanofluids is emphasized in studies related to the thermophysical aspects of nanofluids. In this work, a comprehensive review of the most important theoretical, empirical, and computer-aided predictive models of TC of nanofluids is undertaken. Then, several intelligent models, including multilayer perceptron (MLP), radial basis function neural network (RBFNN) and least square support vector machine (LSSVM) were developed to predict relative TC of nanofluids using 3200 experimental points. The database encompasses 78 different nanofluids, covering extensive-ranged parameters; namely temperature ranging from - 30.00 to 149.15 degrees C, particle volume fraction in the range of 0.01-11.22%, particle size from 5.00 to 150.00 nm, particle TC ranging from 1.20 to 1000.00 W/mK and base fluid TC of 0.11 to 0.69 W/mK. Combining the developed intelligent models into a committee machine intelligence system (CMIS) provided more accurate predictive model. The CMIS model exhibited very low AARE values of 0.843% during the training and 0.954% in the test phase. Moreover, a comparison of performances showed that CMIS largely outperforms the best theoretical and empirical models. Lastly, by performing Leverage approach, the statistical validity of CMIS was confirmed and the quality of the employed data was checked.