Predicting welding deformation of mild steel bead-on-plate joints by means of artificial neural network and inherent strain method

Abstract

The inherent strain method (ISM) is a powerful tool for predicting welding deformation. The implementation of ISM requires the acquisition of inherent deformation data in advance. However, when faced with a practical engineering problem, using conventional methods to obtain this data impromptu consumes too much preparing time. To address this issue, we proposed an idea by integrating thermal-elastic-plastic finite element method (TEP FEM), artificial neural network (ANN), and ISM to predict welding deformation efficiently. To implement this idea, Q355 steel bead-on-plate joints were selected as research objects in the current study. TEP FEM was used to compute a series of inherent deformations under the conditions with various heat inputs and plate thicknesses, and some corresponding experimental mock-ups were conducted to verify the simulation accuracy. The simulated results were then used to train and validate an ANN model with the output layer of inherent deformation components. The trained ANN model serves as a dynamic database to provide the required inherent deformation data instantly, and the ANN-ISM approach was applied to predict the welding deformation efficiently. Besides, the influences of heat input and plate thickness on inherent deformations of bead-on-plate joint were investigated. The out-of-plane deformation modes of the joints with different plate thickness were examined.