Modeling of Residual Stress in Butt Welding

Abstract

Welding is one of the major manufacturing processes widely used in all industries because of its inherent properties of joining similar or dissimilar metals efficiently and economically. Residual stresses are inherent, unavoidable detrimental and have significant effect in welded structures. Researchers have developed many techniques to predict welding residual stress. Intelligent tools and techniques have been applied to predict residual stresses to meet the demands of automation in industries. Existing tools are limited in application and needs attention. This research paper addresses the development of finite element model and neurohybrid models for the prediction of residual stress in butt-welding. Residual stress model is developed by the finite element method (FEM). Data sets from FEM model are used to train the developed neuro-based hybrid models such as neural network model trained with genetic algorithm (NNGA), neural network model trained with particle swarm optimization (NNPSO), and neural network model trained with fuzzy system, adaptive neuro fuzzy inference system (ANFIS). Among the developed models, performance of ANFIS model is superior in terms of computational speed and accuracy. Developed models are validated and reported. These developed models find scope in welding shop floor environment to set the initial weld process parameters.