Contact resistance prediction of zirconium joints welded by small scale resistance spot welding using ANN and RSM models

Abstract

The total electrical resistance between electrodes is a major factor in small scale resistance spot welding process uncertainty. The initial resistance between electrodes is depending on the current profile, electrode force and dome radius of hemispherical electrodes. A comparison between the experimental results of the expected resistance between electrodes and the predicted values of using ANN and RSM methods is the main goal of this study. E110 zirconium alloy with a thickness of 0.25 + 0.25 mm and 440 MPa tensile strength is used. An increase in dome radius of hemispherical electrodes reduces mean resistance values for zirconium alloy. The rate of preheating current rise has no appreciable effect on stabilization of resistance between electrodes in all cases. Stepwise current rise significantly reduces dispersion of resistance values for zirconium alloy. However, their dispersion significantly decreases after preheating in comparison with initial values. The empirical model predicted by the response surface methodology (RSM) is compared with the backpropagation algorithm model of artificial neural networks (ANN). On zirconium specimens, the absolute maximum error in RSM technique was 9.92% while 4.43%in ANN. Furthermore, in most cases the absolute maximum error in ANN is lower than RSM.