Prediction of angular distortion in the fiber laser keyhole welding process based on a variable-fidelity approximation modeling approach

The angular distortion is one of the most common types of distortions frequently observed in laser weld assembling processes, which leads to a decline in welding joints' quality and additional costs of rework. Therefore, it is of great importance to control and reduce the welding-induced angular distortion by selecting appropriate welding process parameters. The challenge is how to predict the welding-induced angular distortion in the whole process parameter design domain accurately and efficiently. To address this challenge, a variable-fidelity approximation modeling approach is developed in this paper, where two different levels of fidelity data are integrated for predicting the angular distortion in the laser welding process. In the proposed approach, a three-dimensional thermo-mechanical finite element model is developed as a low-fidelity model, while the laser welding experiment is taken as a high-fidelity model. A low-fidelity radial basis function (RBF) model is constructed based on the sample data from the finite element simulation. Then a linear tuning strategy is introduced to bring the low-fidelity RBF model as close as possible to the data from the laser welding experiment. Finally, the variable-fidelity approximation model is constructed by adopting a scaling function to calibrate the remaining differences between the tuning low-fidelity approximation model and the high-fidelity data. Two types of validation approaches are adopted to compare the prediction accuracy of the variable-fidelity approximation model with those of the single-fidelity approximation models solely constructed with laser welding experiment or finite element simulation. Results illustrate that the prediction ability of the developed variable-fidelity approximation model outperforms those of the single-fidelity approximation models.