August 2026 Volume 8

AUTOMATION

Artificial intelligence can help forgers improve quality, consistency, uptime, and energy performance — but only when it is applied to clearly defined operational problems. Bryan DeBois, Director of Industrial AI at RoviSys, explains where manufacturers should begin, why plant-floor AI is different from ChatGPT and how companies can separate real opportunities from hype. BEYOND THE BUZZ: A PRACTICAL PATH TO AI ON THE FORGING FLOOR By Bryan DeBois and Angela Gibian

A rtificial intelligence has quickly become one of manufacturing’s most discussed technologies. The term appears in nearly every conversation about automation, workforce development, operational efficiency, and the factory of the future. But when forging leaders hear “AI,” Bryan DeBois believes they should spend less time thinking about the technology itself and more time thinking about the decisions being made throughout their plants. “Where could AI help the plant make a better decision faster, more consistently or with less dependence on tribal knowledge?” DeBois said. That question shifts the conversation away from broad promises about digital transformation and toward the challenges forgers confront every day: inconsistent processes, unplanned downtime, scrap, rework, energy consumption, and the loss of experienced employees. The value of AI in forging is not necessarily found in a chatbot. It may come from using plant-floor data to recommend better process setpoints, applying computer vision to inspect parts or process conditions, or eventually allowing an AI system to participate in carefully defined operational workflows. In each case, the technology should support a measurable manufacturing objective. Not All AI Is the Same One of the first steps for manufacturers is understanding the difference between Generative AI and Autonomous AI. Generative AI tools such as ChatGPT are designed primarily to work with language and knowledge. They can summarize reports, draft procedures, answer questions, organize information and help employees locate knowledge more efficiently. DeBois describes these applications as knowledge problems in the “carpeted space” of a business. Those tools can be useful for manufacturers, but they are not designed to directly control a press, furnace, robot, or material handling system. Plant-floor problems are operational, and the consequences of a poor recommendation can be significantly greater. Autonomous AI is intended to operate within a defined industrial context. Instead of drafting a document, an autonomous system might recommend a billet temperature adjustment before a press cycle. It could suggest changing press stroke timing based on temperature and load feedback or slowing a material-handling sequence when upstream conditions move outside the normal operating window.

“The key difference is that Generative AI helps people solve knowledge problems, while Autonomous AI can recommend or take action in a defined operating context,” DeBois said. That does not mean manufacturers should immediately give AI control over production equipment. In most early applications, the system should advise operators, engineers, or maintenance personnel rather than act independently. Strengthening Engineering Judgment The distinction between Generative and Autonomous AI becomes especially important in a safety-sensitive manufacturing environment. A poorly written sentence in an office document can be corrected. A poor recommendation involving a forging press, furnace, robot, or hot material can create a far more serious risk. AI used on the plant floor must therefore be bounded, monitored, and introduced with the same discipline manufacturers apply to automation, control systems and safety systems. “In forging, AI should be used to strengthen engineering judgment, not substitute for it,” DeBois said. The safeguards should reflect the potential consequences of each decision. At a minimum, manufacturers should establish clear operating boundaries, human review requirements, version control, auditability, cybersecurity protections and performance monitoring. Employees also need to understand when they are receiving an AI-generated recommendation, what information the system considers and how much confidence they should place in its output. For higher-risk decisions, AI should remain advisory until the system has been tested under actual operating conditions and approved through the company’s engineering and safety processes. Start With Existing Plant Problems The strongest AI opportunities in forging are often connected to familiar operational priorities rather than futuristic concepts. DeBois recommends looking first at processes that are repeatable enough for the system to learn from but variable enough that skilled employees regularly make judgment calls. Potential applications include: • Heat treatment and furnace performance

• Press and hammer behavior • Die life and tooling condition • Lubricant application

FIA MAGAZINE | AUGUST 2026 31