A comparison between the back-propagation and counter-propagation networks in the modeling of the TIG welding process

The use of neural networks to model tungsten inert gas (TIG) welding is explored in this paper. Both the back-propagation and counter-propagation networks are used to associate the welding process parameters with the features of the weld-pool geometry. A comparison between the back-propagation and counter-propagation networks in the modeling of the TIG welding process is given. It is shown that both the back-propagation and counter-propagation networks can model the TIG welding process with reasonable accuracy. However, the counter-propagation network has better learning ability for the TIC welding process than the back-propagation network, although the back-propagation network has better generalization ability fur the TIG welding process than does the counter-propagation network. (C) 1998 Elsevier Science S.A.