Computer vision sensing and intelligent control of welding pool dynamics

This paper presents a successful application of intelligent technologies, such as computer vision, image processing fuzzy-neural networks, for sensing, characterization, modeling and control of the welding pool dynamic process in the pulsed Gas Tungsten Arc Welding (GTAW). The image processing algorithm is discussed in detail. The results of the 2-dimesion and 3-dimesion geometric characterization of the weld pool from the single-item pool images are presented. Neural models and self-learning neural controller for the weld pool dynamic process have been developed. The experiment results on the process indicate that the designed vision sensing and control systems are able to emulate a skilled welder's intelligent behaviors: observing, estimating, decision-making and operating. The work in the paper shows great potential and promising prospect of artificial intelligent technologies in the welding manufacturing.