August 2026 Volume 8

AUTOMATION

Separating Opportunity from Hype For FIA members evaluating AI, DeBois’ central recommendation is straightforward: Start with use cases, not technology. Identify decisions that affect quality, throughput, uptime, energy performance, or safety. Select a focused problem, establish a baseline, test the technology, and measure the result. Manufacturers should also involve a trusted system integrator early, particularly when a project requires information from multiple control, production, quality, and maintenance systems. “These projects succeed or stall based on three things: choosing the right AI for the use case, connecting the right data from the right places, and tying the results to measurable business outcomes,” DeBois said. The most valuable industrial AI applications may not resemble the dramatic, fully autonomous factories often featured in technology predictions. They are more likely to appear as incremental improvements: an earlier warning, a more consistent setpoint, a better maintenance priority or a recommendation that helps an operator avoid a recurring defect. Hear more from Bryan DeBois at the FIA Fall Meeting of Members, October 19-21, 2026. Register online at www.forging.org/events.

In that situation, the first AI project is not building an algorithm. It is building the data foundation that will eventually allow the company to use one. Turning Tribal Knowledge into a System AI also offers a potential method for preserving the knowledge of experienced operators, engineers, and maintenance professionals. That knowledge cannot simply be extracted from a database. It must be taught. DeBois describes machine teaching as a process in which subject matter experts help define what an AI system should observe, what good and bad decisions look like, and which operating boundaries it must respect. In a forging operation, an experienced employee might explain how billet appearance and temperature affect the next process step, how subtle changes in press behavior can indicate a developing problem, or how furnace response varies under different production conditions. The goal is not to document every thought an operator has. It is to convert expert judgment into a structured decision framework that can be applied safely and consistently. This process also keeps employees at the center of AI development. Operators and engineers are not merely users of a finished system; they help teach the system how the process actually works. A Practical First Pilot For a small or midsize manufacturer, a strong first AI pilot should be narrow, measurable, and connected to a meaningful operational problem. The required process and quality data should already exist or be practical to collect. The team should also have known outcomes against which the AI recommendations can be compared. A pilot might focus on:

• Reducing scrap for one product or part family • Recommending billet temperature targets • Detecting abnormal furnace performance • Improving lubricant application consistency • Identifying process drift before final inspection

• Adjusting furnace setpoints to reduce energy consumption The project does not need to transform the entire plant to be successful. A pilot that generates a measurable improvement and teaches the organization how to collect, connect and use data has created a foundation for future projects.

FIA MAGAZINE | AUGUST 2026 33