APPLICATION OF ARTIFICIAL-INTELLIGENCE TECHNIQUES TO RESISTANCE SPOT-WELDING

Despite decades of research, controlling the spot welding process is a difficult task, particularly with the increased use of zinc coated st eels. Conventional control techniques have been limited by the fact th at many weld quality monitoring signals are prone to changes in their profile as a result of electrode wear, leading to misinterpretation of the signals. Current stepping has been introduced to maintain consist ent weld quality by increasing the welding current in predetermined st eps to compensate for reduced current density with the growth of the e lectrode tip. However, the optimum current stepping programme is diffi cult to establish in practice. To overcome the present limitations of spot welding control systems, greater effort needs to be applied to de velop intelligent control systems. Artificial intelligence (Al) techni ques, relying less on mathematical process representation, may be idea l for controlling this highly non-linear process. The development of a neural network based process model for weld size prediction is descri bed. Taguchi techniques have been used to establish the optimum networ k parameters, and weld nugget diameter has been classified from electr ical welding data and information regarding the condition of the weldi ng electrodes. Based on the results of the investigation and a recent review, it is felt that a combination of the neural network, with its mapping and pattern recognition capabilities, and a fuzzy logic contro ller, with its ability to handle vague and imprecise data, is likely t o offer greatest benefits in overcoming the limitations of existing co ntrol systems.