Prediction of transverse and angular distortions of gas tungsten arc bead-on-plate welding using artificial neural network

• The GTA bead-on-plate welding has been studied by numerical and experimental methods.
• The results between BPN model and experiments show a good agreement.
• The BPN model developed in this paper shows good efficiency and reliability.

The angular distortion and transverse shrinkage are often generated in gas tungsten arc (GTA) bead-on-plate welding process, which leads to additional costs of rework. Therefore, it is beneficial to estimate the welding deformations prior to bead-on-plate welding in terms of several process parameters. This paper presents the development of a back propagation neural (BPN) network model for the prediction of angular distortion and transverse shrinkage generated in GTA bead-on-plate welding process. The model is based on the results from finite element (FE) simulations. The GTA bead-on-plate welding for S304L stainless steel was simulated using finite element method, and experiments were conducted to verify the accuracy of the FE model. The experimental results were also used as testing samples for the BPN model. Welding speed, current and voltage were considered as the input parameters and the angular distortion and transverse shrinkage were the output parameters in the development of the BPN model. The correlation coefficients and percentage errors for all the samples were calculated to evaluate the prediction accuracy of BPN model. The results show that the BPN model developed in this study can predict the angular distortion and transverse shrinkage with reasonable accuracy.