The welding process was monitored with a XIRIS XVC-1000 welding camera and standalone software built by TWI software engineers using the XIRIS Source Development Kit (SDK). An example of the camera set-up is shown in Figure 1.
The software used brightness thresholding to identify different regions of an image, within two user-defined areas of interest (AOI). This works by identifying regions of an image that have digital brightness values above a user-defined value. The software then measures the height and width of this region in terms of pixel size, and this is then converted to a physical measurement by use of a known physical measurement as a scaling factor. An example image of the software is shown in Figure 2.
The software was used to study the weld pool width, arc width and arc height, as these were considered likely significant control parameters for the welding process that could be varied by a mechanised system. Weld pool length is unfortunately difficult to measure with a trailing camera system due to the presence of the arc and the parallax effect, so a side-view camera would be necessary as an additional monitoring system. This shows the difficulties of replicating the relatively free human movement around an arc performed during welding.
The software recorded the above measurements throughout the welding process, allowing time-based values of these measurements to be examined. For example, Figure 3 shows measurements taken during two welding runs. Figure 3a shows a "standard" welding condition, with no change in the welding parameters during the process. The measurements remain consistent once a steady state weld pool is established. Note, because this is a pulsed welding process, the overall "brightness" of the image varied between a low and high value, synchronised to the pulse current. Two methods of eliminating this double counting were considered. The first was a thresholding operation within the software that only records values above a certain limit. The second was synchronising the camera to the current, such that images are only taken during the pulse cycle. Off the shelf hardware exists to perform this.
Figure 3b shows the measurements recorded from a welding run in which the arc length was varied during the process from a short arc length, to a long arc length, and back again. The effect of this on the welding process can be seen. During the lengthening of the arc length, the weld pool increases in width up to a maximum value as the arc spreads, at which point heat transfer becomes less efficient. A similar effect can be seen in the arc width, as it spreads on lengthening. The arc length was measured precisely on the pulse element of the cycle, but shows no effect on the background element, as during this part of the cycle, the brightness threshold was focused around the tungsten electrode.
These measurements were further synchronised to the electrical parameters of the welding process, to determine the delay between changes in welding parameters and image features. An example is shown in Figure 4, which shows welding image measurements synchronised to welding current and arc voltage, in which the welding current was increased and in particular, the weld pool width increased as a result. As seen in Figure 4, the delay between change in welding parameters (of 5-10%) and change in image measurement was of the order of 0.5 seconds.