Predictive analytics of oil-based non-newtonian nanofluid's viscosity with multi-layer perceptron neural networks

Nanoparticle addition in a base fluid known as nanofluid is being applied extensively in today's technology due to its superior thermal and viscous properties. However, experimental studies on new nanofluid combinations to determine their thermophysical properties require ample cost and time. Hence, artificial neural networks are suggested in this research. This study developed two multi-layer perceptron (MLP) neural network models to predict the viscosity of two different oil-based non-Newtonian nanofluids, i.e., ZnO-Coconut oil- and Cu-Gear oil-based nanofluids. This viscous property was chosen as the output variable of the ANN models due to its remarkable effects on heat transfer and fluid flow. The viscosity of nanofluid depends on various factors such as temperature, nanoparticle concentration, and shear rate. Therefore, These three parameters were chosen as the models' input variables. Experimental data was obtained from the existing studies, and machine learning algorithms were applied to predict viscosity. For each nanofluid, 14 network architectures were established by varying hidden layers and number of neurons to find the optimal topology of the model. Statistical parameters such as R2, MSE, RMSE, and MAPE were used to evaluate the performance of the models. Results indicated that the evaluation criteria values obtained for neural network models signified that the developed models could predict viscosity values accurately. The ANN-predicted outputs showed an excellent agreement with the actual experimental data values.