Innovation Fall 2026

Novarc product manager Romario Gumulia, P.Eng., says Novarc has accumulated 850,000 hours of welding video in its library. P hoto : W endy D P hotography

One mode of particular interest for Smith is the physical technique of the welder holding the torch. "We're trying to capture the actual motion of the welding gun while the welder is welding, then use that to train algorithms," Smith said. He thinks that current, more mature, vision techniques could be significantly augmented by data about the movement of the welder. “I don’t believe a purely vision-based approach is really capturing welder skill,” he said. “It’s not really capturing the way the welder manipulates a puddle.” Ultimately, these various forms of data collection support a potential future with greater end-to-end autonomy. “What I would like to see is a small system capable of learning from a welder and agile enough to maneuver into confined spaces, then implement some of what it’s learned,” said Smith. “We probably won’t have enough data for it to do everything, but at least we’ll have a framework. The next step would be to capture enough data to build upon that framework.” Expertise to guide autonomous systems As systems evolve and grow capable, new forms of expertise will need to evolve to adapt to these advancements in welding with and alongside machines. "I expect a strong global focus on giving traditional automation the ability to perceive its environment and adapt in real time without human intervention," said Gumulia. Welders and weld inspectors, however, remain accountable for ensuring quality welds and safe machine operation. “Any time we look for a process improvement, we have to develop a complete welding procedure for it,” said Banduyang. “Any time we are looking for ways

to improve the procedures, swap wires or change the consumable brand, we get involved again.” This means every machine and new procedure comes with a new set of controls. “We cannot use the procedure we have for semi-automatic or hand-built for automated machines or robotic machines,” said Moallem. “So whenever we are implementing a new machine, we have to develop a new weld procedure and qualification record for that; that’s why it absorbs the majority of our time.” At manufacturing facilities like Seaspan, the process of integrating automation involves dozens of professionals, and typically involves extensive deliberation from teams across various departments to assess the value, impact, and checkpoints for quality control. Procedure qualification tests, witnessed by relevant classification societies, are required for individual welds on any new process, as well as submissions for destructive and nondestructive testing. Future welders will need to understand a variety of subjects deeply, complementing advances in automation. "If you're a junior person today, it makes a lot of sense to learn about the technology. Learn automated welding; become an expert not just in welding itself but in doing it with robots and advanced systems," said Van Dyck. "Welding is an extremely complex, nonlinear process," he added. "There's a huge amount to learn: material science, arc physics, puddle behavior, process parameters, and so on. The way I see it, the future is about solving those challenges one by one with robotics and automation."

30

Fall 2026

Innovation