A real-time passive vision system for robotic arc welding

Visual sensing in arc welding has become more and more important. However, it has remained a challenging problem due to the harsh environment, especially the interference of the extremely strong arc light. A typical industrial solution for seam detection is to use laser scanner which however still suffers from several limitations. Solutions for weld pool sensing mostly rely on high speed cameras and powerful external illuminations and therefore the wide use in industrial environments is limited. In this paper we present a real-time passive machine vision system for both seam tracking and weld pool sensing in robotic arc welding. The system, which consists of a CCD camera and a carefully-selected narrowband pass filter, is capable of extracting the geometrical profile of the seam that is close to the arc region and the weld pool boundary simultaneously in real time. This visual information can be further utilized by the control system to navigate the welding robot and adjust the welding parameters. Experiments including several real-time demonstrations have been made and verified the feasibility of the proposed vision system.