OPTIMUM DESIGN BASED ON MATHEMATICAL-MODEL AND NEURAL-NETWORK TO PREDICT WELD PARAMETERS FOR FILLET JOINTS

The welding process variables of welding current, are voltage, welding speed, gas flow rate, and offset distance, which influence weld bead shape, are coupled with each other but not directly connected with wel d bead shape individually. Therefore, it is very difficult and time co nsuming to determine the welding process variables necessary to obtain the desired weld bead shape. Mathematical modeling in conjunction wit h many experiments must be used to predict the magnitude of weld bead shape. Even though experimental results are reliable, prediction is di fficult because of the coupling characteristics. In this study, the 2( n-1) fractional factorial design method was used to investigate the ef fect of welding process variables on fillet joint shape. Finally, a ne ural network based on the backpropagation algorithm and an optimum des ign based on mathematical modeling were implemented to estimate the we ld parameters for the desired fillet joint shape. Mathematical modelin g based on multiple nonlinear regression analysis was used for modelin g the gas metal are welding (GMAW) parameters of the fillet joint. It was shown that the neural network and optimum design for estimating th e weld parameters could be effectively implemented, which resulted in little error percentage difference between the estimated and experimen tal results.