Investigation and modeling of the SMAW coating flux thermal properties using neural network and regression analysis

The present work aims to design and develop shielded metal arc welding electrode coating fluxes. The thermal properties of the electrode coating fluxes are studied. Ternary phase diagrams of CaF2−CaO−Al2O3 and SiO2−BaO−Al2O3 are used to determine the range of constituents. 27 flux mixture combinations were created using an extreme vertices design approach. The corrected optical basicity is estimated and utilized for the theoretical estimation of sulfide capacity and water vapor solubility. Thermogravimetric analysis, and differential scanning calorimetry analysis was performed to quantify the thermal response of the flux while heating. Thermogravimetry analysis shows variation of weight loss from 8.49% to 12.25% with the variation in flux composition. Calcite and witherite has positive influence on weight loss. Differential scanning calorimetry shows an overall change in enthalpy from −1076.73 to −7786 J/g. Hot disk analysis reveals fluorspar has a positive effect, whereas calcite has a negative effect on thermal conductivity. The effect of individual flux constituents and their interactions on thermal properties is discussed using regression analysis. The artificial neural network is used for predictive modeling, and the accuracy of its predictions is compared to regression analysis. In comparison to the regression, the model's mean square error improved by nearly 37% on average. The reduction in the mean absolute percentage error of the unseen test data suggests that the artificial neural network approach has better modeling capability.