AI-powered vision system improves pipe welding automation

Novarc Technologies of North Vancouver, B.C., introduced its Spool Welding Robot (SWR) for pipe welding applications in 2016. Images provided

You might be familiar with collaborative robots, which work alongside humans without the threat of knocking them out as they complete their assigned tasks. If a collaborative robot bumps into a human, it will stop immediately, instead of plowing through the person as if they were an empty cardboard box. In essence, man and machine can “collaborate,” as the human guides the robot’s action or simply works nearby as the robot performs as instructed.

For pipe welding applications, Novarc Technologies, North Vancouver, B.C., introduced its Spool Welding Robot (SWR) in 2016. It was built around a cobot that was designed to be used specifically for pipe fixtured on a rotating chuck. The cobot, armed with a welding gun, moves up and down and forward and back as it performs the weld. Typically, a welder oversees the operation, adjusting the cobot during the welding process if he sees a problem or if the cobot has to contend with sloppy weld preparation, such as a bevel angle that falls outside of expected tolerance windows. The SWR can weld flanges, T-joints, and elbows.

This type of pipe welding traditionally has been the domain of skilled welders. The high-mix/low-volume work called for talented welders who could smoothly go from one job to the next and handle the variances in pipe prep and fit-up. The SWR was designed to alter that dynamic.

“It helps to make the junior welder much more productive,” said Soroush Karimzadeh, CEO, Novarc Technologies. “With the SWR, a junior welder can weld as much as three or four highly skilled welders over an eight-hour time span—and with much better quality.”

Karimzadeh said that metal fabricators and mechanical contractors have reacted positively to the release of the SWR, with Novarc having installed more than 30 systems around the world. But in dealing with the realities of competitive markets, those same early adopters of the cobot pipe welding technology asked what else the automation might be capable of. Would it be possible to have the cobot work more independently without having a welder watching the operation every second? Could the SWR be set up so that an operator could take on other tasks as the cobot was welding or even oversee multiple SWR systems at the same time?

This inspired advancements over the last two years in which Novarc worked on developing deep neural algorithms that could help the SWR learn and adjust on its own, as a human welder or an operator of the basic SWR system would. The result is NovEye, an artificial intelligence-powered vision system that advances the SWR’s automated welding capabilities.

As an example of how the NovEye works, Karimzadeh described how it handles tack welding. When a pipe is prepped for a cobot, a welder uses feathered tack welds instead of bridge tack welds so that the cobot can weld over them. A cobot welding system that doesn’t have the advantage of an AI vision system requires the system operator to alter the system control as the cobot approaches the tack weld. The control changes the welding parameters as the cobot welds over the tack weld, achieving complete fusion. When complete, the operator reengages the normal root weld program. The NovEye lessens that operator engagement because the vision system sees the tack weld and adjusts welding parameters on its own.

“The other part of this is the consistency,” Karimzadeh said. “We noticed that with some operators the inputs could be different depending on who was running the machine or what the skill level of the operator was. The NovEye means a lot more consistency. Our algorithm is going to react the same way and in the most optimal way every single time.”

The initial release of this feature is targeted at mechanical contractors that are pipe welding to ASME B31.9.0 standards. In the ensuing months, the automation upgrade will be released to metal fabricators welding to ASME B31.1 AWS B3D1.1 and B31.3 D1.3 standards.

This is the first of other upgrades that will take place and be sent out to SWR users. Karimzadeh said that pipe welders expect their capital equipment investments to keep pace with technological advancements, especially in the age of digital connectivity.

A new AI vision system from Novarc Technologies aims to automate pipe welding further.

One of the planned updates will focus on weld process monitoring, during which the vision system will be able to track the welding process and provide real-time insight to the operator as to how things are going. Such feedback provides a record of the weld quality, which can be used to determine if the joint is acceptable or if improvement is needed.

“Pipe fabrication is becoming more competitive, especially in this pandemic era,” Karimzadeh said. “We see that there’s a demand for more automation, which might be a good thing. This field already has a big shortage of skilled welders, and more automation can free up welders from doing repetitive work and have them do more difficult welds.”