Challenging ANN and RSM approaches to forecast β-SiC nanoparticles efficacy on performance of liquid ethylene glycol and propylene glycol

Abstract This study challenges the capabilities of artificial neural network (ANN) as well as response surface methodology (RSM) in estimating the viscosity (μ) and thermal conductivity (k) of β − SiC/EG and β − SiC/PG nanofluids. The implementation of these methods is performed by defining statistical criteria with a great obsession to forecast viscosity ratio as well as thermal conductivity ratio with the least error. The nanoparticle of β − SiC has a bigger effect on μEG and μPG so that they increase up to 78.71% and 56%, while kEG and kPG improve under the best conditions up to 14.64% and 4.85%, respectively. Statistical calculations show that for RSM, R-square is 0.987 for μ β − SiC / EG μ EG , 0.975 μ β − SiC / PG μ PG , 0.99 k β − SiC / EG k EG and 0.987 k β − SiC / PG k PG . This figure for ANN is 0.994, 0.992, 0.994 and 0.991. This implies that ANN ability is superior. Although β − SiC nanoparticles increase the viscosity by 78.7% (for EG) and 56% (for PG) which is not desirable, fortunately, fluid mechanics-based calculations reveal that in turbulent regime, the pumping power increases by only 17.2% and 13.5% under worst conditions.