Process parameter optimization for laser-magnetic welding based on a sample-sorted support vector regression

Abstract

Magnetic field assisted laser welding (LW-MF) shows great potential in the jointing of large structures. The quality of the welding joint in LW-MF largely depends on the selection of process parameters. In this study, an integrated process parameter optimization framework is developed for magnetic field assisted laser welding. Firstly, Taguchi method is selected to generate sample points and the LW-MF experiments are carried out to obtain the bead geometrical characteristics. Secondly, a sample-sorted SVR (SS-SVR) metamodeling approach is developed to make full use of the already-acquired prediction error information for fitting the relationships between multiple process parameters and the bead geometrical characteristics. A detailed comparison between the developed SS-SVR metamodeling approach and existing SVR metamodeling approach for prediction accuracy is performed. Then, the particle swarm optimization is used to solve the process parameters optimization problem, in which the objective function values are predicted by the developed SS-SVR metamodel. Finally, verification experiment is conducted to verify the reliability of the obtained optimal process parameters. Results illustrate that the proposed integrated process parameter optimization framework is effective for obtaining the optimal process parameters and can be used in LW-MF for practical production.