Prediction of transverse welding residual stress considering transverse and bending constraints in butt welding

Jeongung Park,Gyubaek An,Ninshu Ma,Seong-Joon Kim
DOI: https://doi.org/10.1016/j.jmapro.2023.07.043
IF: 5.684
2023-07-24
Journal of Manufacturing Processes
Abstract:Welding residual stress is generated by the thermal stress caused by rapid local heating and cooling during welding using a welding heat source. The distribution and size of the residual welding stress are affected by the joint type, welding conditions, constraints, etc., which also affect the fatigue, fracture, and buckling strengths. Therefore, because the magnitude and distribution of residual stress are very important for the safety evaluation of structures, guidelines for structural integrity evaluation suggest a method for deriving the welding residual stress. However, the welding residual stress presented in the guidelines is in a state without constraint; if a constraint exists, it is suggested to use the yield stress of the base material. This can lead to a very conservative design using the yield stress as the welding residual stress in the recent trend of the increasing use of high-strength steels with increased yield stress. Therefore, it is very important to investigate the relationship between constraint and residual stress during welding and to predict the residual stress according to the degree of constraint. In this study, the effects of member thickness, yield stress, lateral constraint, and bending constraint on the welding residual stress in butt welding were investigated, and predictive factors for welding residual stress were established. To this end, a thermal elastic–plastic analysis was performed by changing the thickness, yield strength, lateral constraint, and bending constraint size of the member. Subsequently, the analysis results were used to train a deep neural network (DNN) to predict the residual stress at the center of the weld and toe. In addition, the influence of input factors that affect the prediction accuracy during DNN learning was considered.
engineering, manufacturing
What problem does this paper attempt to address?