Optimisation of artificial intelligence models and response surface methodology for predicting viscosity and relative viscosity of GNP-alumina hybrid nanofluid: incorporating the effects of mixing ratio and temperature

Abstract

AbstractThe viscosity properties of GNP-alumina hybrid nanofluids are of significant importance in various engineering applications. This study compares the predictive performance of response surface methodology (RSM), artificial neural network (ANN), and adaptive neuro-fuzzy inference system (ANFIS) for the viscosity (µrel) and relative viscosity (µrel) of GNP-alumina hybrid nanofluid at varying mixing ratio (0–3) and temperature (15–55 °C). The ANN and ANFIS models were optimised by varying the number and type of neurons and membership functions (MFs), respectively. In contrast, the RSM model was optimised by varying the source model. The efficacy of the models was assessed using various measures of performance metrics, including residual sum of squares, root mean square error, mean absolute error, and mean absolute percentage error (MAPE). The ANN architecture with 4 neurons exhibited exceptional proficiency in forecasting the µnf, achieving an R2 value of 0.9997 and a MAPE of 0.3100. Meanwhile, the best ANN architecture for the µrel was achieved with 5 neurons, resulting in an R2 of 0.9817 and MAPE of 0.2588. Furthermore, the ANFIS model with the difference of two sigmoidal MFs and the product of two sigmoidal MFs for µnf and Generalized Bell MFs for µrel exhibited the best performance with (3 5) and (4 5) input membership functions, respectively. An R2 value of 0.9999 and 0.9872, with a corresponding MAPE value of 0.0945 and 0.1214, were reported for the optimal ANFIS architecture of µnf and µrel, respectively. The RSM model also produced its most accurate prediction with the quadratic model for both µnf and µrel, with an R2 value of 0.9986 and 0.8835, respectively. Thus, comparative analysis across various models indicated that the ANFIS model outperformed others regarding performance metrics for both µnf and µrel. This study underscores the potential of ANN and ANFIS models in accurately forecasting the viscosity properties of GNP-alumina hybrid nanofluids, thus offering reliable tools for future applications.