Prediction of laser welding qualities of Al alloys using regression and machine learning techniques

Abstract This work compared different machine learning models such as linear regression, polynomial regression and XG-Boost for the prediction of laser welding qualities in aluminum alloys. The key weld quality parameters are ultimate load, weld width and penetration depth. Each model was trained and validated based on data experimentally collected by varying laser power, scanning speed and offset distance to compare them. Quantitative results are shown to prove that XG-Boost produces a better predictive accuracy, as it gives a root mean square error (RMSE) of 0.05 for ultimate load, 0.03 for penetration depth, and 0.02 for weld width. On the other hand, its linear regression counterpart has higher values of 0.08, 0.06, and 0.05, while polynomial regression-clearly outperforming its linear variant-averaged about 0.04 in these metrics. While this is so, high R-squared values, predicted by the XG-Boost model across ultimate loads, are indicative of its better competency in the capturing of complicated patterns, especially with regard to data outliers. These findings confirm the capability of XG-Boost to perform precise parameter optimization in laser welding by significantly reducing experimental trial needs and helping manufacturing efficiency with reliable data-driven predictions.