Intelligent assessment of effect of aggregation on thermal conductivity of nanofluids-Comparison by experimental data and empirical correlations

Improving the performance of energy-based systems is a key factor for potential applications of nanofluids. Nanoparticle aggregation is a complicate phenomenon that changes thermal conductivity of nanofluids and their heat transfer performances. There's available no accurate correlation for accounting effects of aggregation on thermal conductivity of nanofluids over wide ranges of operating conditions. Therefore, we focus on designing accurate artificial intelligence (AI) approaches for prediction the effects of aggregation on thermal conductivity of different nanofluids. 1065 experimental data were collected from 36 different literatures and used for designing the AI-based approaches. Thermal conductivity of aggregated/non-aggregated nanofluids is related to temperature, nanoparticle size, aggregation size, nanoparticle concentration in base fluid, and synthesis procedure of nanofluid. Different AI-based paradigms including multilayer perceptron (MLP), radial basis function (RBF), generalized regression (GR), and cascade feedforward backpropagation (CFB) were examined and their performances were compared. Statistical analyses confirmed that the CFB neural network with eight hidden neurons outperforms all other considered models. Both computational effort and accuracy margin revealed that the Levenberg-Marquardt is the best training algorithm. This model predicted the experimental data with absolute average relative deviation (AARD) of 1.37%, mean square errors (MSE) of 4.75 x 10(-4), and coefficient of determination (R-2) of 0.98452.