Optimized ANFIS models based on grid partitioning, subtractive clustering, and fuzzy C-means to precise prediction of thermophysical properties of hybrid nanofluids

Abstract

Applying machine learning algorithms in the prediction of nanofluids' thermophysical properties such as density, viscosity, thermal conductivity (TC), and specific heat capacity (SHC), can lead to cost and time reduction in practical applications. The present research aimed to accurately predict the thermophysical properties of water-based oxide-MWCNT hybrid nanofluids by adopting a whole search strategy for structural/training optimization of the ANFIS models with different types of clustering techniques, including grid partitioning (GP), subtractive clustering (SC), and fuzzy c-means (FCM). To evaluate the optimized ANFIS models various statistical criteria, ARD-based pie charts, MOD plots, violin graphs, and well-known theoretical/experimental correlations were employed. The results revealed that the optimal GP-ANFIS model performed better than other ANFIS approaches in modeling the nanofluid's specific heat capacity (R = 0.99992 and MAPE = 0.0359%) and thermal conductivity (R = 0.99833 and MAPE = 0.2177%). Also, the optimal SC-based ANFIS approach presented the highest precision model for the nanofluids density (R = 0.99886 and MAPE = 0.0369%) and viscosity (R = 0.99887 and MAPE = 0.4206%). The sensitivity analysis indicated that inputs of nanoparticles density, solid volume fraction, nanoparticles SHC, and nanoparticles TC are the most influential parameters in predicting nanofluid density, viscosity, specific heat capacity, and thermal conductivity, respectively.