Visual Characters Extraction Of Weld Pool Based On Three Path Meseurement During Aluminum Alloy GTAW Process

Pulse GTAW Welding is often used in arc welding of aluminum alloysheet because of its stability. During aluminum GTAW welding，in order toachieve the integrative information of current, voltage, sound, images andother signals, our laboratory designed a system to collect weldinginformations. The purpose of this paper is to study the method of weldingpool visual information extraction.Due to the strong heat-sensitive of aluminum alloy, experiencedwelders are required to observe the molten pool during the manual welding,to obtain the pool size, penetration and weld seam situation, etc. Nowadays,the welding industry is getting more and more automated and intelligent.We use visual sensing system instead of human eyes to observe thewelding pool. Passive visual sensing image acquisition method hasincomparable advantage in the simulation of human eye observation. As alot of parameters are needed during the pulsed GTAW aluminum welding,a three-path visual sensing system is used to obtain the images of thefront-top, front-back and the back paths in the same picture at the sametime and get more complete welding information.In this paper, the method of visual attention is used in pool imageprocessing. First， the paper introduced on visual attention, chose atop-down visual attention modeling method based on the characteristics ofimages obtained from the experiments, processed only the interested areasin the weld pool image to obtain the parameters of the welding process thatare most needed, and established a flowchart of visual sensing imageprocessing methods. The general method of image processing is also usedhere, with the results of visual attention for comparison. Since only a smallvisual attention is processed using the visual attention method， this method can be more clear and efficient，especially in real-time detection.Welding defects occur during the welding process. The featurecharacters of porosity, slag, non-penetration, weld leak and linearmisalignment are all different. An additional bottom-up visual attentionmodel based on image saliency is added to process the unexpected features.According to different criteria of defects...