Machine learning and non-equilibrium properties of nanofluids

Accurate prediction of the non-equilibrium thermophysical properties, such as viscosity and thermal conductivity of nanofluids, is crucial for the efficient design and operation of engineering systems. This study explored deep learning neural network models to predict the transport properties of 10 nanofluids. Two artificial neural networks were developed: the first is a four-layer network with 13 neurons in each hidden layer and five inputs, while the second is a three-layer network with four input parameters and nine hidden neurons. Molecular parameters were derived from a semi-empirical equation of state for nanofluids. The key innovation of this study is the application of deep learning to the development of these neural networks. The models demonstrated remarkable performance, achieving an average absolute relative deviation of 2.90 % for viscosity and 1.05 % for thermal conductivity. Additionally, the accuracy of the deep neural networks was compared with two semi-empirical models for these properties.