Statistical Characterization of Stress Concentrations along Butt Joint Weld Seams Using Deep Neural Networks

In order to ensure high weld qualities and structural integrity of engineering structures, it is crucial detect areas stress concentrations along seams. Traditional inspection methods rely on visual manual geometry measurements. Recent advances in the field automated measurement techniques allow virtually unrestricted numbers inspections by laser measurements profiles; however, compare different welding processes manufacturers, a deeper understanding statistical distributions seams required. Hence, this study presents an approach statistically characterize types butt joint For purpose, artificial neural network created from 945 finite element simulations determine concentration factors at joints. Besides higher quality predictions compared empirical estimation functions, new can directly be applied all welded including arc- laser-welded joints, coupled with 3D-measurement devices. Furthermore, sheet thickness ranging 1 mm 100 assessed.