Data Assimilation in the Welding Process for Analysis of Weld Toe Geometry and Heat Source Model

Abstract

Finite element simulations are widely conducted to evaluate the heat transfer and deformation during welding. Basically these welding simulations require input variables such as shape parameters and heat source parameters, which are not directly measured by the experimental method. In this study, two methods were proposed to obtain these input parameters more efficiently: a method of automatically identifying toe radius and reinforcement angle from height profile, and a method of estimating a heat source model in welding simulation. In the first method, the toe radius and reinforcement angle were extracted from the height profile by Akaike’s information criterion. The extracted results were consistent with the manual fitting results. In the second method, the optimal combination of the heat input parameters was automatically searched by Bayesian optimization. Comparing the accumulated regrets, it was found that the probability of improvement and upper confidence bound provide more efficient optimization than the other acquisition functions in the calibration of the heat input parameters. Both temperature history and shape of fusion zone and heat-affected zone calculated at the optimized condition were in good agreement with the experimental results. These results demonstrated that the two proposed methods are effective to create a numerical model for welding simulation.