Physical AI: The Welder’s Apprentice?

Caren Dripke, head of robotics development at Lorch Schweißtechnik GmbH, compares AI-enabled welding cobots to having an apprentice with you when you’re welding. 

“It's like having an apprentice, and it's not his very first day on the job, but rather, he has seen a couple of things. And you only need to point out the general description of the job, and the apprentice can figure the rest out by himself,” said Dripke. 

With all the hype around artificial intelligence (AI) these days, it’s easy to lose sight of the fact that AI is already being used across multiple industrial applications. This includes automated welding, where collaborative robots (cobots) with AI capabilities are being increasingly deployed to improve ergonomics for skilled welders and drive efficiency. 

AI comes in different flavors and is intended for different purposes. Some systems guide you through the programming process, for example, while others provide joint tracking and path-planning capabilities. A common thread running through all AI welding tools is that they virtually eliminate programming as a barrier to the adoption of automation. 

Production Bottleneck: Robot Programming   

According to David Pietrocola, cofounder and CEO of Cohesive Robotics, robot programming is a major bottleneck in welding automation projects, especially for small- to medium-sized businesses. Cohesive Robotics’ Argus OS allows welding cobots to perform common weld joints and tack welds with automatic vision-based detection out of the box. 

“Some manufacturers don't have a manufacturing engineer on staff, let alone an automation engineer. The no-programming aspect, the AI, the vision — these smarter systems are helping to sidestep those bottlenecks and are helping companies bring in robotic technology quicker,” said Pietrocola. 

A welding cobot with Cohesive’s 3D camera mounted on its arm takes millions of data points from different positions as the robot moves, and the company’s Argus OS processes this data using AI running on a dedicated, embedded GPU. 

“It understands where the parts are. It knows what it's looking for. And using that information, it's able to program the robot to go where it needs to go,” explained Pietrocola. 

Drawing from Real-World Deployments 

Hirebotics’ AI-powered Beacon software provides suggestions designed to guide end users of all experience levels through the automated welding setup process. 

“It's like having a copilot,” said Zach Boyd, CTO at Hirebotics. “For first-time users of automation, it really helps to get them over that initial hurdle. The welders are still in control, but certain things are simplified so they can focus on more of the craft, while the cobot focuses on the more monotonous tasks.” 

Leveraging AI to optimize parameters such as voltage, wire feed speed, and travel speed, the system uses contextual data to suggest settings tailored to the specific application and material. 

The suggestions Beacon provides are not drawn from a limited list of pre-determined weld settings and recipes, but from hundreds of real-world deployments of cobots that have been analyzed by AI. Moreover, if end users encounter a problem, they can take a picture of the welding cell, and Beacon will use AI to suggest fixes. 

Where complex traditional programming requires users to think through multiple waypoints and input many details, AI-powered tools like SeamPilot remove much of this complexity, according to Lorch’s Dripke. 

“You don't have to provide many input parameters. You don't have to set up all the torch angles. You don't have to provide every single waypoint. Rather, you just say, ‘Here are the parameters, here's the seam [joint], please weld this.’ SeamPilot takes care of all the geometrical paths. That's a huge time saver, because waypoints usually take the most programming time, so you can easily weld right away,” explained Dripke. 

Automatically Adjusts Travel and Work Angles 

Meanwhile, Vectis Automation has developed a range of embedded AI features designed to lower barriers to welding automation deployments. These features reference points trained by the cobot operator, automatically adjusting the travel and work angles to match production standards, and auto-detecting corner transitions to enable optimized welds on complex geometries. 

Instead of having to precisely place the torch and manually teach all points, these AI features perform the calculations for the operator. Additional features such as intelligent multipass welding, automatic joint tracking, and smart path generation for plasma cutting are also available. 

Finding a welder with years of experience who knows, for example, the best practices for running a certain work angle and travel angle for a specific weld is a real challenge for metal shops, states Josh Pawley, vice president of business development and founding partner at Vectis Automation. 

“AI weld path optimization makes it possible for a new welder to quickly jump in and meet these standards, and smooths some of the ‘tribal knowledge’ barrier that welders naturally gain over the course of a long career. We’ve even seen dozens of times where the knowledge of an experienced welder in the shop is essentially passed on to the next generation via the cobot tool,” said Pawley. 

Automating HMLV Production 

Programming traditional welding robots takes considerable time, which can be a barrier to automating high-mix/low-volume (HMLV) production runs. Depending on the size of the job lot and how long it takes to set up your welding robot, it may not seem worthwhile to automate the welding process at all. 

Acccording to Dripke, by drastically reducing setup times, AI-powered technologies enable job shops to confidently deploy automation even on smaller jobs. 

“When your robot is easy to program, it tips the scale toward small lot sizes being easily automated. SeamPilot figures out many things on its own, so operators don’t have to. This easy programming enables high productivity even on small lot jobs that previously would have been too time-consuming to set up for automated welding,” said Dripke. 

The Importance of Being Human 

AI can handle a lot of welding tasks, but keeping a human in the loop makes for even more effective solutions, says Dripke. 

“The human is the most important resource in welding. It’s important to have somebody who understands different materials, welding parameters, and job requirements. You also need a human to check the weld quality after the fact. Automating welding as a whole process is not easily done, especially with the weld details; relying on the experience of welders is the way to go,” explained Dripke. 

So, it's not a case of ‘either-or’ when it comes to humans and cobots. In fact, skilled welders who can also operate welding automation are highly valued and very well paid. 

“When you find a skilled welder, that's great. If you find somebody who can program robots, that's great too. But if you find somebody who has skills in welding and can program a robot, you have most likely found somewhat of a unicorn,” said Dripke. 

But with major industrial economies worldwide suffering a dramatic skilled labor gap, AIs that fully automate deployment and welding processes are probably inevitable, says Cohesive’s Pietrocola. 

“The U.S. is short hundreds of thousands of welders. It’s a similar story in Europe and China. We see our technology as a way to supercharge your existing welding workforce, to increase production. And with an aging welding workforce, systems like ours enable workers to actually work longer if they choose, because they're not the ones who will necessarily be welding all day, every day,” explained Pietrocola. 

In some future cases, fully automated welding driven by AI may well be the norm, but cobot-based welding is still about human-robot collaboration, says Hirebotics’ Boyd. 

“There’s a lot of hype about AI and what it can do. However, at this point, humans are still better at some tasks, so it’s very much still a collaboration between a human and a machine. And beyond the hype, there are also really practical examples, like Beacon, of being able to embed AI into software that helps our customers get started quicker, without it being overly intrusive,” said Boyd. 

This article was written by Will Healy III (director of product and industry marketing at Teradyne Robotics, North Reading, Mass.) for the American Welding Society.