Finite element analysis, prediction, and optimization of residual stresses in multi-pass arc welding with experimental evaluation

Prediction and reduction of unwanted tensile Residual Stress of welded stainless-steel plates is presented in this paper. Validated finite element analysis and Artificial Neural Network (ANN) is employed to simulate and mathematically model the process, respectively. Taguchi design of experiments tool is utilized to generate input data for finite element analyses and also to choose the most accurate ANN structures. RSs are minimized using three methods: Taguchi suggestion, Comprehensive factorial search, and Particle Swarm Optimization, whose accuracy and response pace increases and decreases respectively in this order. Furthermore, adding and removing extra weld lines was proposed to reduce unwanted residual stresses by up to 50%. Finally, the shapes and amounts of results are experimentally verified using contour method and proposed novel application of roughness testing. Micro-grain structures of the welded samples were also investigated, and RSs were discussed considering metallography images.