Abstract
In this study, the accuracy of applying artificial neural networks and response surface methodology in estimating k(SiO2/EG) was examined. Thermal conductivity prediction in the temperature range of 5-65 degrees C and mass fractions of 0.005-5 wt.% was performed. Considering the constraints of minimizing the mean square error (MSE) as well as maximizing the R-squared value, the most appropriate polynomial based on linear regression (for RSM) and the optimal number of neurons were obtained through performing the least squares methodology. Statistical calculations showed that MSE value for ANN and RSM techniques was 0.0000342 and 0.0001577, respectively. Comparing the R-squared values of ANN (0.993) and RSM (0.968) methods, it was found that from this perspective, the artificial neural network is superior. Comparing the results of the simulation and the laboratory, it was observed that both methods are not very accurate at low mass fractions. However, the potential of the ANN technique was greater than RSM one. Also, the usefulness of SiO2/EG nanofluid through helically coiled tube heat exchanger (HCTHE) was challenged from the perspective of the second law of thermodynamics. Performing exergy balance revealed that the exergy destruction is intensified at higher mass fraction and lower temperature. Applying RSM on irreversibility affirmed that the cubic linear regression led to statistical criteria of R-2 = 0.9999 and MOD less than 0.008%, and therefore, it is recommended to navigate the exergy destruction through HCTHE.