Predictive ANN modelling of Thermorheological properties of Iron-Oxide yield stress nanofluid

Abstract

The intent of the research is to find the dependency of the volume fraction of nanoparticle (φ) and the temperature on the absolute viscosity (µnf) of Fe3O4 nanoparticles in Carbopol polymer gel. Rheological and stability analysis of the solution is identified. A total of 48 viscosity values has been calculated from experiments using two different base fluid concentrations and two different nanofluid concentrations at eight different temperatures. The data gathered are used for the training of an ANN (Artificial Neural Network) to observe results in a predefined range of two input criteria. It uses a feed-forward perceptron ANN with a temperature input, a volume concentration input, and a viscosity output. The topology was established by trial and error, and the two-layer model having ten neurons in the hidden layer that used the tansig function produced the best results. Ten training functions were utilized to analyze the best result for nf prediction, and the trainbr algorithm was found to be the best ANN. Due to the trained ANN, the anticipated value of viscosity is obtained from each temperature and volume concentration combination. The best results were witnessed with trainlm algorithm with an MSE value of 5.92e-4 and a R2 value of 0.9988 for forecasting of viscosity. Nanoparticle volume concentration increases with viscosity, while temperature increases cause viscosity to decrease. As the temperature rises from 15°C to 50°C, the shear stress value drops with a corresponding shear rate. The shear stress value of the associated shear rate decreases as the nanoparticle concentration rises.