MULTI-OBJECTIVE NEURAL NETWORK MODELING FOR IMPROVING STUD ARC WELDING PROCESS JOINING

An effective stud weld algorithm has been developed using a tensile strength response. A multi objectives neural network modeling was used to get higher tensile strength with lower variance. A multi-layer perception (MLP) was trained by back propagation Levenberg-Marquardt (LM) [8: 16: 2] algorithm using sample testing measures. The result was compared with Taguchi experimental design method. The comparison shows that neural network is an effective method to represent this multi objective target problem. The verification of results shows that the ultimate tensile strength increases by 30.84% and the standard deviation around mean reduces by 30.06%. The results prove that neural network is a powerful tool to represent the multi level eight input parameters with two objectives first is to maximize the ultimate tensile strength and the second is to minimize the standard deviation of tensile strength with almost negligible error (4.96 x10(-10)) for representing data.