Estimation of laser weld parameters using surrogate modelling technique

Abstract

Abstract A surrogate technique based on Gaussian Process (GP) is used for predicting quality of laser in case of laser welding process that may be supported by induction preheating. FEM-based solution of the problem is computationally expensive because it combines computation of 3D coupled nonlinear electromagnetic and temperature fields. The quality of laser welds is quantified with weld depth, which depends on a number of input parameters. The paper deals with two of them —thickness of the welded steel sheet and power of the laser beam. First, selected FEM simulations allow finding data describing the dependency between the two input parameters and weld depth. These data allow creating a surrogate model that is able to predict weld depth at any point close to the points where the results are known. The principal goal is to essentially save the computational time. The surrogate model also allows estimating prediction plausibility and running the full FEM calculation in cases where the prediction is not sufficiently accurate. The methodology is illustrated with a typical example whose results are discussed.