Content area
Full Text
Humanitarian military applications
1 Introduction
Nowadays, more and more welding robots have been applied in the automatic manufacturing field. Generally, most welding robots serving in practical production remain at the level of teaching and playback, which must be taught in advance for plenty of time as well as taught and programmed again for another different workpiece as they cannot self-rectify the offset during the welding process. So it is very important for welding robots to have the ability to adjust themselves and process some sensor information autonomously, which is important in practical production to realize autonomous welding ([1] Chen, 2007; [2] Dilthey and Stein, 1992).
There have been many studies on visual sensing techniques for observing weld seam and weld pool images because they have non-contact sensing capability and can provide abundant information. In the factory, the active vision sensor had been effectively applied in weld seam tracking. [7] Kovacevic and Zhang (1997) researched the image sensing of the weld pool surface by high-speed camera synchronized with laser sensor in Tungsten inert gas. [3], [4] Fridenfalk and Bolmsjö (2003, 2004) used both the arc sensing and laser scanning method to track the six-dimensional weld seam. Comparing the active vision sensor, the passive vision sensor has more abundant information, so it can realize to track complex curve seam. In recent years, passive visual sensor was always a focus in the field of the sensing technique used in the welding robots. [8] Kuo and Wu (2002) used the fuzzy logic based on the passive vision to realize the weld seam tracking. [13] Kim et al. (1995) used the vision technique to track the weld seam in robotic arc welding. [16] Zhou et al. (2006) proposed an approach for autonomous acquisition of seam coordinates for arc welding robot. It should be pointed out that the above passive sensing methods can only track the plane weld seams, by which the height information of the image can be hardly got. Through the method of the binocular matching, the height information of the image can be computed. But the processing time cycle is long and the accuracy is low. Here, arc sensing was adopted. Many scholars ([6] Kodama et al. , 2005; [10] Nakamura and Hiraoka, 2004; [11], [12] Pan,...