Improving the prediction accuracy of thermal finite element analysis for laser welding through an automated optimization method

This paper proposed an automated optimization method to improve the prediction accuracy of thermal finite element analysis for laser welding. The method used particle swarm optimization (PSO) to find the optimal heat source parameters with the minimum prediction error, and the PSO was improved with a boundary mutation strategy of attenuation reflection. To improve the optimization speed, the effects of model geometry and latent heat on the prediction accuracy and solution speed of thermal analysis were studied. The results showed that the reduced 3/50-length model could increase the solution speed with a speed-up of 137 times and had the same predicted results as the full model. The latent heat could increase the nonlinearity, and thus gave rise to a speed-down of 4 times in solution speed but a decrease of 6.06% in prediction error. Therefore, a reduced 3/50-length model considering the latent heat was used in the optimization. The parameter optimization was first performed in the thermal analysis using a welding speed of 2.4 m/min, and then the optimized optimal parameters were used at other 5 speeds. The results showed that the error function of the parameters was a multi-peak function. In addition, it could be seen that the maximum, minimum, and mean errors of the weld cross sections were 9.47%, 4.36%, and 6.81% respectively, while the errors using the trial and error method were 12.48%, 7.20%, and 9.98% respectively. This indicated that the proposed method had a speed-up of about 6 times and higher prediction accuracy.