<title>Neural network model for predicting the backside dimension of weld pool during pulsed GTAW process</title>

Abstract

Pulsed GTAW was used widely in butt welding of thin plate. Top surface depression occurred without filler wire in full penetration, while reinforcement height was assured with filler wire. Currently butt welding process control of thin plate welding during pulsed GTAW with filler wire was depended on manual experience and the consistency of seam shape was hardly attained. Based on self-developed vision sensor, double-side images of weld pool were captured simultaneously in a frame. By image processing the topside dimension and shape of weld pool, such as area, length, maximum width, the similarity of reinforcement, and the coefficients of multinomial regression of boundary, and the backside dimension such as area, length, maximum width and the similarity of height were calculated. A fractional factorial technique was used to design the experiment. Artificial neural network was applied to establish the steady model for predicting backside dimension of weld pool. The input of the model was the topside dimension, shape of weld pool and welding parameters, such as pulse current, base current, arc voltage, pulse duty ratio, welding speed, and wire feeding rate. The output of the model was the backside dimension of weld pool. Finally the variance method was used to test the validity of the model.