Improving joint hardness in vibration-assisted robotic gas metal arc welding of dissimilar AISI 304 stainless steel and AISI 1018 mild steel through ANN coupled with GA optimization

Abstract

Abstract The paper focused on optimizing vibration-induced robotic gas metal arc welding (GMAW) to enhance Rockwell hardness of the fusion zone (FZ), stainless steel heat affected zone (HAZ-SS) and mild steel heat affected zone (HAZ-MS) of the dissimilar AISI 304 stainless steel and AISI 1018 mild steel joint. Five input parameters each with three levels were employed for the execution of 32 experiments designed using response surface methodology central composite design. The experimental investigation revealed that maximum FZ hardness of 107.6 HRB was achieved at current 100 A, voltage 26 V, welding speed 7 mm sec −1 , gas flow rate 6 L min −1 , and amplitude 0.4 mm; and maximum HAZ-SS and HAZ-MS of 86.8 HRB, and 71.9 HRB respectively were obtained at current 85 A, voltage 22 V, welding speed 5 mm sec −1 , gas flow rate 8 L min −1 , and amplitude 0.6 mm. A hybrid artificial neural network-genetic algorithm (ANN-GA) method was utilized to predict and optimize the response variables and the related control parameters. Best prediction of the ANN model was attained at an MSE of 0.0011376 and coefficient of regression R of 0.9988 for normalized data and an MSE of 2.797784 on original data for a 5-10-3 configuration back-propagation network using the Bayesian Regularization algorithm. The genetic algorithm optimal results for FZ, HAZ-SS, and HAZ-MS were 119.2 HRB, 91.3 HRB, and 78.6 HRB respectively. From the confirmation test, an average response of FZ 116.4 HRB, HAZ-SS 90.3 HRB, and HAZ-MS 77.5 HRB were achieved with error percentages of 1.11%, 2.35% and 1.42% respectively.