DESIGNING A FEEDFORWARD NEURAL NETWORK FOR WELDING CONTROL IN OIL AND GAS INDUSTRY EQUIPMENT.

This paper presents a method to process the welding-arc voltage with neural network. The authors describe a WIG (Wolfram Inert Gas) arc welding industrial control system based on the correction of the welding torch trajectory through the welding seam. The welding trajectory is corrected by using a control system in which the control function is realized via a feed-forward neural network trained with the back-propagation algorithm. The feedback signal for the neural controller is obtained from an acquisition system used to get the voltage values from the arc-welding process. Six training algorithms are taken into consideration and compared: for each algorithm the error of the welding head is shown. [ABSTRACT FROM AUTHOR]