A data-driven physics-informed neural network for predicting the viscosity of nanofluids

Abstract

Nanofluids have been applied in various fields, such as solar collectors, petroleum engineering, and chemical engineering, due to their superior properties compared to traditional fluids. Among the various thermophysical properties of nanofluids, viscosity plays a critical role in thermal applications involving heat transfer and fluid flow. While several conventional machine learning (ML) techniques have been proposed to predict viscosity, these conventional models require many experimental measurements to be optimized and make accurate predictions. This study reports a novel ML method using a multi-fidelity neural network (MFNN) to accurately predict the viscosity of nanofluids by incorporating the physical laws into the model. The MFNN correlates a low-fidelity dataset derived from the prediction of the theoretical model with a high-fidelity dataset, which consists of experimental measurements. It is shown that the MFNN can recover the rheology of nanofluids and outperforms the conventional artificial neural network due to incorporating the underlying physics of nanofluids into a model.