Using an optimized RBF neural network to predict the out-of-plane welding distortions based on the 3-2-1 locating scheme

Document Type : Article


1 Department of Mechanical Engineering, Ferdowsi University of Mashhad, Mashhad, Iran

2 Ferdowsi University of Mashhad, Department of Mechanical Engineering, Mashhad, Iran


This study deals with the effect of locator positioning in 3-2-1 locating scheme to control the out-of-plane distortion in gas tungsten arc welding of sheet metals. To apply this locating scheme on the sheet metals, a suitable fixture is designed. The distortion of the welded plates has been predicted using the radial basis function (RBF) neural network. To gather the experimental data employed in the RBF modeling process, a set of welding tests is performed on the sheet specimens by varying the positions of the three locators. The parameters of the network are optimally selected using the simulated annealing (SA) optimization algorithm. The average and maximum error computed for the test dataset were respectively 2.43% and 5.30% while in some cases the error falls below 1%. The results of the RBF network show very good agreement with the experiments and it can be concluded that this modeling technique can be utilized successfully in predicting the welding distortions when the 3-2-1 locating scheme is used.


Main Subjects