Parameter Optimization in Resistance Spot Welding of AISI 1060 Steel Using Adaptive Neural Fuzzy Inference System and Sensitivity Analysis

Resistance spot welding process of AISI 1060 steel has been experimentally investigated by studying the effects of welding current, electrode force, welding cycle and cooling cycle on tensile-shear strength. Using the response surface methodology, experimental tests are performed. An adaptive neural-fuzzy inference system is applied to model and predict the behavior of tensile-shear strength. Additionally, the optimal parameters of adaptive neural-fuzzy inference systems are obtained by the gray wolf optimization algorithm. For modeling the process behavior, the results of experiments have been employed for training (70% of data) and testing (30% of data) of the inference system. The results show that the applied network has been very successful in predicting the tensile-shear strength and the coefficient of determination and mean absolute percentage error for the test section data are 0.96 and 6.02%, respectively. This indicates the considerable accuracy of the employed model in the approximation of the desired outputs. After that, the effect of each input parameter on tensile-shear strength is quantitatively evaluated with the Sobol sensitivity analysis method. The results show that the tensile-shear strength of the joint rises by increasing the welding current and welding cycle and also decreasing the electrode force and cooling cycle.