Welding Vision Systems. . . .by Nello Zuech, Contributing Editor

One of the earliest application-specific implementations of 3D-based machine vision was in welding. This application of structured light or Z-axis range data collection more or less evolved alongside the use of similar approaches in the wood products industry, where volume-based cutting optimization demonstrated significant gains in yield. In the case of welding, the application of machine vision was driven by the aerospace market where safety considerations are of paramount importance and cost of failure extremely high.

Some of the earliest work in welding vision was conducted at Marshall Space Center or funded by them at places like Ohio State University and the Edison Welding Institute. More or less at the same time similar work was being conducted at Canadian-based research centers as well as research centers in the UK, Europe and Japan. 

Welding machine vision systems are the basis for fully automatic welding stations be they robotic-based or flexible automation centers. These systems do not require constant observation and control adjustments. Such systems are well suited to large volume part production. The automotive and off-road vehicle industries are major users of machine vision assisted automatic welding stations.

As noted, welding vision systems are generally based on structured light or range data collection. In the case of structured light-based systems, laser diodes are used to project a pattern of light at a preset angle in the shape of a single line, multiple lines or a cone of light yielding a projected circle. Triangulation mathematics determines the Z-axis data.

In the case of the range data arrangement, a laser light sweeps across the joint of the part(s) being welded and the time it takes the laser to reflect to the 3-D camera head determines the location of the joint. In both cases, the systems eliminate all wavelengths of light associated with the welding process by using filters that only pass the wavelength of light associated with the lasers used. Also, camera optics with varying field of view accommodate a range of working distances, seam types and sizes.

In the case of robot welders, the use of welding vision provides compensation for lack of precision in the robot manipulator, part fixturing and tooling and the workpiece itself. Programming the robot to follow complex weld joint paths is also simplified. The result is an adaptive welding station. In some cases just finding the seam in the first place to start the welding process can be an issue.

Today one finds welding vision systems that can find the seam compensating the initiation of the weld cycle for positional errors. The welding vision system eliminates the positional uncertainty of the start point, locating often to 25 microns or better. These same systems can sometimes be programmed to monitor specific points along the seam as well as the end of seam position to assure weld completeness. In addition to position, at each measurement point these 3D systems also monitor width and depth to control welding parameters. Systems that use a projected circle or multiple lines can also calculate trajectory.

There are also welding vision systems that perform full seam tracking continuously feeding positional information back to the robot to compensate in real-time for any positional variables. These systems essentially move the camera in sync with the welding torch over the joint. Generally the requirement is not only to assure the welding torch tracks the joint, but also to assure the joint is properly filled using the required number of passes of the welding torch. By monitoring width and depth, these systems can also provide feedback to compensate for welder parametric variables in real-time as well as dimensional variations stemming from thermal distortions created by the welding process itself.

Some of these seam trackers operate in advance of the weld operation and some through the arc. In tracking, the welding speed and look ahead distance are used to calculate the vertical and horizontal corrections to keep the welding torch over the seam. In general these systems can operate at welding speeds in the 3 – 4 m/minute range, some are now available capable of up to 25 m/minute. Positional accuracies are typically 0.1mm. Versions of these systems can also monitor the weld bead profile essentially behind the torch as well as perform defect detection.

Tracking systems are interfaced to a robot or welding machine PLC using analog, digital or serial interfaces. In the case of welding machines the positional corrections are made via motorized slides. Most of the commercially available welding vision systems can interface to any number of commercially available robots: ABB, Cloos, Comau, Fanuc, KUKA, Motoman, Nachi, Neos, etc.

The camera optics are generally protected from the smoke and spatter of the welding process by a shield with an inexpensive disposable plastic window. This has to be changed regularly when the window becomes covered with the spatter. In some cases, the camera head is cooled by air or welding gas to keep the temperature down. The cooling air can also act as an air curtain to keep the contaminants off the disposable window.

Not all welding vision systems are equal. Some are more suited for welding thin materials, some for thick materials, some for pipe and tubing. It is also the case that some are better suited for certain welding approaches: MIG (gas metal arc), TIG (gas tungsten arc), SAW (submerged arc welding), plasma, laser, etc. Some are also more suited for welding complex geometries than others. The type of image processing and feature extraction and analysis performed by the welding vision system may depend on the joint type and features like sheet metal edge, gap, mismatch, vertical and horizontal position, orientation, etc. Hence, when investigating welding vision systems it is critical to understand the application requirements.

As noted, versions of these systems have been adapted to immediate post weld inspection at the same rate as the weld joint is fabricated. These systems verify the geometric profile of the bead and can identify certain defect conditions based on predefined criteria: mismatch, gap, bead width, convexity/concavity, weld slope, undercut, burn-through, root concavity, excess penetration, and pinhole.

Welding vision systems are a proven technology in spite of the harsh environment. The payback from using welding vision is real: weld quality is generally better and more consistent, higher productivity, lower production costs, less environmental impact, reduction of scrap and rework, etc. Another consideration is the need in many companies to combat the shortage of skilled manual welders.

Automating post weld inspection eliminates the typical inspection inconsistencies and misinterpretations experienced when people perform the inspection as well as delays (as people cannot perform quality inspections at the rates the joints are being welded). It can also lead to elimination of the acceptance of substandard welds or the unnecessary repair of welds that are acceptable. Another benefit is that such systems offer the ability to observe trends and implement process improvement programs accordingly. Another byproduct could be the ability to archive digitally both the image of the weld as well as the corresponding data collected.