A NEURO-FUZZY APPROACH TO SELECT WELDING CONDITIONS FOR WELDING QUALITY IMPROVEMENT IN HORIZONTAL FILLET WELDING

Many researchers have developed algorithms to control welding paramete rs for a desired weld bead geometry. Unstable welding conditions induc e an unsound bead, resulting in poor mechanical properties at welded j oints. Generally, the dominant variables affecting weld bead geometry are welding current, are voltage, and welding speed. In practice, it i s difficult to determine the proper combination of welding conditions because of excessive nonlinear and complex characteristics of welding processes. The relationship between welding conditions and weld defect s cannot be easily represented by mathematical models, and it is diffi cult to predict weld bead geometry resulting from welding conditions. A fuzzy rule based method and neural network method are proposed: the neural network method predicts welding conditions appropriate for the desired weld bead geometry, and the fuzzy rule based method chooses ap propriate welding conditions for avoiding weld defects such as undercu t and overlap in horizontal fillet welding. Performance of the propose d neuro-fuzzy system was evaluated through experiments, which showed t hat the system can effectively check and adjust welding conditions in regard to weld defects.