Welding is a widely used industrial process and a subject of extensive research, particularly regarding stress concentration caused by weld geometry. Although several models have been developed to estimate the service life of welded assemblies, their predictions often suffer from significant uncertainty, resulting in a wide dispersion of fatigue life estimates.

This paper presents an experimental methodology aimed at building a representative dataset of welds, for which multiple geometric features are measured, including 3D scanning. These measurements enable the evaluation of stress concentration factors through finite element analysis along the full length of each weld. A regression analysis is then performed on the collected data to derive an explicit model capable of estimating the stress concentration factor with acceptable accuracy based solely on weld geometry.