AI, machine learning, and the future of metal fabrication

On the first day of the 2024 Fabricators and Manufacturers Association Annual Meeting, held in Clearwater Beach, Fla., Gene Marks, speaker and columnist for Forbes magazine, pointed at an eye-opening chart tracking the cost of computer processing speed over the past few decades.

During the FMA Annual Meeting, a Navy SEAL turned leadership consultant brought up an idea that, for those who never served in the military, seemed a bit surprising: decentralized command. “Not only does decentralized command allow you to grow into a role, but as a leader, it allows you to step back and look at the big picture. There’s no way I can have that view if I’m constantly making decisions for my team.”

That was veteran Carlos Mendez, a consultant with Texas-based Echelon Front. His insight went against the popular view of the military, shaped by movie scenes of sergeants screaming at subordinates. The reality is that soldiers can find themselves cut off from central command, and if they don’t have the training or authority to think and act independently, they can be in a world of trouble.

Lives might not be at stake in the fab shop, but livelihoods certainly are. Most metal fabrication occurs in high-product-mix environments. With equipment and software juggling hundreds of jobs, some unexpected variables are bound to throw a wrench into the workday. Fabricators work to minimize the exceptions, but there will always—always—be exceptions.

On the first day of the late-February conference, held in Clearwater Beach, Fla., Gene Marks, speaker and columnist for Forbes magazine, pointed at an eye-opening chart tracking the cost of computer processing speed over the past few decades. “The price of processing speed today is roughly one-one-hundred-millionth of what it was in the 1970s. The fastest computers in 1993 could perform less than 1,000 operations in a millisecond. That’s gone up to over a billion operations. That’s every millisecond.”

The extraordinary power of modern computing has created all sorts of AI tools, but they’re not “total” solutions. They can write an email, design a presentation, and automate certain email tasks. They help immensely in a thousand different ways, but they only get you 80% to 90% there. Humans still need to bring work over the finish line.

This scenario might reflect life on the shop floor one day, though we’ve got a ways to go. As several attendees from custom fabricators discussed during the conference breakout sessions, the challenge is data. Conventional wisdom has it that manufacturers are swimming in it—but how good is that data? Machines and software capture incredible amounts. But in most fab shops, not every machine is automated, and a lot has to happen before and after each manufacturing step. Instead of paper job travelers, operators now use laptops, tablets, even their phones, but they’re still probably keying in job information manually into an ERP system. What exactly happens at a specific work center between that initial clock-in and final clock-out often just isn’t captured.

This sometimes leads to big surprises when fabricators integrate Industrial Internet of Things (IIoT) platforms. These can reveal just how little “real” uptime machines have—that is, when machines are actually cutting, bending, and welding, and good parts are actually being produced. Usually, it’s a fraction of the time people assumed. IIoT is revealing low-hanging-fruit improvements (material staging, standardizing procedures and work practices between shifts, etc.), but it’s also showing that, no matter how dialed-in an operation becomes, exceptions will exist. More so than in past years, discussions at the FMA Annual Meeting really focused on how to best manage them.

Lean principles entered the fray during the conference sessions—like optimizing machine utilization but not at the expense of a plant’s overall throughput. Less-talked-about inefficiencies also entered the debate. During one breakout session, Caleb Chamberlain, co-founder of OSH Cut (and fellow columnist for The Fabricator), brought attendees through the customer experience he and his team designed. OSH Cut doesn’t make to print. In fact, it has no prints at all, which raised some eyebrows in the audience. Customers upload design files directly to the OSH Cut website, which performs a design-for-manufacturability (DFM) analysis. If there’s an issue, the customer can make changes and then upload the design again. From there, nesting, machine programming, and myriad other order-prep tasks all happen automatically.

OSH Cut isn’t the only fabricator to do this. A few in the U.S. offer similar services, and Europe has a collection of “web shops,” 247TailorSteel, a Dutch operation, being the most well known. The model won’t work everywhere, but it does bring up larger questions about what activities in the metal fabrication supply chain truly add value, and where those activities (especially DFM) should happen.

Brian Steel, CEO of Cadrex Manufacturing Solutions and a panel participant at the conference, represented the other end of the metal fabrication spectrum. After acquiring a multitude of plants, Cadrex is now one of the largest contract metal fabricators in the country. The company also has adopted software that runs factory-wide simulations, weighs various production options, then suggests what should work best.

Fabricators and Manufacturers Association Annual Meeting attendees took a deep dive into operational excellence, skill, and what will define future success: good systems, good data, and, most important, good employees.

All this reveals the increasing importance of software innovations, the best of which are aiming to weigh the effects of thousands of variables in high-product-mix manufacturing and, ultimately, help skilled people make better decisions. Software won’t account for everything, which brings the importance of problem-solving and decentralized command to the fore. The last thing FMA Annual Meeting attendees want is to lead an automated shop where software makes all the decisions and people just mindlessly do what they’re told.

Operators shouldn’t avoid using new technology or change machine programs or tools just because that’s what they prefer. But they also shouldn’t run a machine program or job that truly doesn’t work. They need to be able to identify what truly is an “exception,” then have enough knowledge and authority (supported by good systems and procedures) to act and get the job done. No matter what the future of software, machine learning, and AI looks like, employee skill and curiosity will remain a fabricator’s key competitive advantage.