Image processing for measurement of three-demensional gas metal arc weld pool surface: Algorithms are explored for processing the image of a grid laser pattern reflected from the dynamic GMA weld pool surface

Observing and measuring the weld pool geometry in the gas metal arc welding (GMAW) process is the key to understanding and control of the complex welding process, and to improving and validating numerical welding models. In this paper, a novel machine vision system to measure the highly dynamic three dimensional front surface geometry of GMAW weld pool is proposed. A grid pattern is projected onto the weld pool surface. The projected laser is reflected from the specular weld pool surface and the reflection is intercepted by an imaging plane. A high-speed camera fitted with a narrow band optical filter is adopted to capture the corresponding reflected pattern, which is deformed during the reflection. To utilize the reflected pattern to reconstruct the weld pool surface, the correspondence of the reflection points with the projected pattern must be established. This paper addresses the algorithm to first evaluate the captured images, segment and identify reflection pattern from series of captured images, identify projected lines and establish line-to-line correspondence. The establishment of the correspondence provides the data needed to reconstruct the weld pool surface based on the law of reflection.