Prediction of nanofluid viscosity using multilayer perceptron and Gaussian process regression

Abstract

More than a decade, a numerous experimental and theoretical studies of thermophysical properties of nanofluids are conducted to reveal its heat transfer characteristics. Due to nanofluid unique thermal properties, it is broadly used in various applications from automobile applications to biomedical applications. Despite that various experimental and theoretical studies of nanofluids are developed, the accordance between them is very little and also it is tiresome and expensive. To predict the thermal properties in an easy way, soft computing tools are utilized. In this research work, dynamic viscosity ratio of Al2O3/H2O is predicted using machine learning techniques like multilayer perceptron and Gaussian process regression. In the proposed multilayer perceptron—artificial neural network model, varying a range of neurons in the hidden layer and using Levenberg–Marquardt as training function, it is found that 6 neurons in the hidden layer give less root mean square error value of 0.01118. Different kernel functions are opted to train the proposed Gaussian process regression model, and it is found that Matern kernel function shows the best performance with less root mean square error value of 0.018, and regression coefficient value of both the models is 0.99. This research work will reduce the experimental test run cost, and the models are accurate in prediction.