Computational intelligence based prediction of joint penetration in laser fabrication

Abstract In the present study, laser welding of stainless steel sheets of 2.5 mm thickness was conducted to study the weld geometry parameters. The inputs considered here for laser welding were beam power, welding speed, and beam inclination angle. Experimental runs were conducted based on the central composite full factorial design. A multivariate regression model was developed to establish interdependency between input and output of laser welding. The novelty of the paper is in implementing an adaptive neuro-fuzzy inference system (ANFIS) for prediction of weld bead parameters in laser welded joint. An evolutionary algorithm like genetic algorithm (GA) and particle swarm optimization (PSO) had been applied to fine-tune the ANFIS parameters and to enhance the prediction capability of the established network. The performances of these networks were evaluated on the result of coefficient of correlation and root mean square error. Coefficient of correlation (R and root mean square error were found to be 0.9636 and 0.0521, respectively for PSO-ANFIS model. The prediction accuracy along with the performance criteria for the developed models pointed out that PSO-ANFIS model have better prediction accuracy than other discussed models.