Automatic Tungsten Inert Gas (TIG) welding using machine vision and neural network on material SS304

Abstract

Welding is a process of joining two or more substances that are based on the principles of diffusion processes, resulting in unification on the materials to be joined. The strength of the weld joint is determined by several parameters, including the weld bead width and the penetration. The width of the weld bead especially the upper part can be determined by looking directly through the CCD (Charge-Coupled Device) camera. But it is difficult to observe the back bead width directly since in practice it is impossible to install the CCD camera at the bottom of specimen. In this paper, Tungsten Inert Gas (TIG) Welding with the welding speed is controlled by the microcontroller for the purpose of adjusting the back bead width has observed. The back bead width is estimated based on data of weld bead width obtained from machine vision, welding speed, and currents that used in this experimental. It's used to obtain a series of data which would have conducted as initial experiments to train and build the neural network system. Results showed that the back bead width is 3 mm on the current 55 A, 60 A, and 65 A have an average error of each current of 0.11 mm, 0.09 mm, and 0.12 mm.