Welding Journal - May 2011

Observation of the weld pool provides an effective path to understand and control the welding process. Observation data also attribute to the validation of numerical models, which helps researchers to analyze and evaluate the welding process. Extensive research has been conducted to provide in situ monitoring of the weld pool. The arc sensor was one of the first few sensors that was adopted for weld pool observation. X-ray (Ref. 1), infrared, and vision-based sensors were also developed and implemented. The weld pool surface is the most direct source of information for human welders to obtain feedback. Skilled welders estimate the welding process by visually observing the weld pool and adjust welding parameters accordingly. In recent years, machine vision has been more and more widely adopted to monitor the welding process. Early efforts focused on extracting a two-dimensional weld pool boundary by directly viewing the scene under the welding gun. Richardson (Ref. 2) developed a coaxial arc weld pool viewing system for the gas tungsten arc welding (GTAW) process. Kovacevic et al. (Ref. 3) used a high-power auxiliary illumination laser to suppress arc radiation. In recent years, weld pool observation evolved from two-dimensional (2-D) boundary extraction to three-dimensional (3-D) surface geometry measurement. Stereo vision (Ref. 4) and shape from shading (SNF) algorithms (Ref. 5) are proposed to reconstruct 3-D weld pool geometry from directly captured weld pool images. However, one of the most challenging tasks that direct-viewing vision-based monitoring techniques must overcome is the interference of arc radiation, which degrades the quality of captured images. The radiation from the arc body is much stronger than that radiated/reflected from the weld pool surface and workpiece. A scene, including weld pool and welding arc, contains a lot more variations than can be captured by a high-speed camera. The majority of the dynamics on the captured image is assigned to the bright arc body, which is not the area of interest. The dynamics of the weld pool area are significantly reduced and cause an inevitable loss of information. The most common techniques that were adopted to reduce arc influence included blocking the arc body using a mechanical method (Ref. 1), using optical filters to observe the weld pool surface at a specific wavelength, and using an auxiliary light source, such as laser (Ref. 3), to suppress arc influence. However, according to current reports, the dynamics of the captured weld pool surface are still not satisfying. Arc light reflected from the weld pool surface is another issue direct-viewing methods have difficulty in overcoming. In the case of weld pool fluctuation, bright spots, which are a result of arc light reflection, are observed in certain local areas on the weld pool. The reflection is significantly brighter than the average weld pool intensity on the captured image, which degrades the dynamic range of the captured weld pool area. When the weld pool surface is reconstructed using stereo vision or SNF method, surface geometry is calculated from captured texture on the weld pool images. The existence of arc light reflection inevitably degrades the weld pool texture and reduces reconstruction accuracy. Reflection-based weld pool monitoring methods, which take advantage of the specular nature of the weld pool surface, successfully reduced arc light interference. Pioneering work was conducted at the UniSUPPLEMENT TO THE WELDING JOURNAL, MAY 2011 Sponsored by the American Welding Society and the Welding Research Council