August 2026 Volume 8

AUTOMATION • Quality inspection • Energy optimization

Condition-based maintenance has been used in industry for decades, but many plants have struggled to implement it broadly. AI can help by analyzing operating data, identifying abnormal conditions and translating raw information into clearer maintenance priorities. Rather than claiming that the press will fail in 12 days, for example, a system could identify a change in vibration, load, temperature or cycle behavior that deserves inspection. Maintenance teams can then combine that warning with their own experience and knowledge of the equipment. The result is a more informed maintenance decision, not an unsupported prediction. Build the Data Foundation AI cannot provide reliable recommendations without reliable information. A manufacturer needs enough data to describe what happened, when it happened and the operating conditions surrounding the event. That may require connecting information from programmable logic controllers, historians, quality systems, computerized maintenance management systems and production records. The data does not have to be perfect. It does, however, need to be understandable and consistent enough to support decisions. Missing timestamps, inconsistent equipment tags and disconnected spreadsheets make it difficult to determine whether two events occurred during the same production cycle. They also make AI recommendations harder for employees to trust. For that reason, many industrial AI projects begin with an AI-readiness assessment rather than immediate model development. The assessment examines what information is currently available, where it is stored, whether the systems can communicate and whether the data contains the necessary process context. In some plants, improving the data infrastructure may produce benefits even before an AI model is introduced. Better access to operating, quality and maintenance information can help employees investigate downtime, compare production runs and identify recurring problems. Legacy Equipment Can Still Participate Older equipment does not automatically prevent a manufacturer from using AI. Every industrial facility operates some degree of legacy technology, DeBois said. The age of the machine is less important than the information available from it. A 50-year-old press with years of reliable operating, maintenance and quality records may be more prepared for an AI application than a newer machine with limited historical context. Manufacturers should begin by determining what data the equipment already produces. Existing programmable logic controllers, sensors or control systems may capture valuable information even if that information is not currently being stored or analyzed. Where useful data does not exist, the company may need to add targeted sensors, collect production information or establish consistent maintenance and quality records.

• Material handling • Process consistency The most successful projects generally begin with one specific problem. Instead of launching a companywide initiative to “implement AI,” a forger might focus on reducing scrap for one product family, optimizing billet temperature before a press cycle or adjusting furnace setpoints to reduce energy consumption without affecting quality. “The best opportunities usually start with a very specific operational pain point, not with a generic AI initiative,” DeBois said. This problem-first approach also makes it easier to evaluate the results. A manufacturer can compare scrap rates, energy use, production output or equipment performance before and after the pilot instead of relying on vague measures of technological progress. Detecting Process Drift Earlier Quality problems in forging rarely result from one variable. Part quality may be influenced by billet temperature, material condition, tooling, equipment behavior, process timing, operator actions and variation introduced during earlier production stages. An AI model that looks at only one signal is unlikely to provide meaningful insight. AI becomes more useful when it can evaluate combinations of process conditions and identify patterns associated with defects, dimensional variation or abnormal scrap. The system might recognize that a particular combination of temperature decline, load behavior and cycle timing frequently occurs before a quality issue. That information can alert employees to investigate before the defect becomes visible during final inspection. The objective is not to replace established inspection or quality control processes. It is to give production and quality teams an earlier indication that the process may be drifting away from its normal operating range. To do that successfully, the system must have enough context to understand what acceptable production looks like. Rethinking Predictive Maintenance Predictive maintenance is frequently promoted as one of AI’s most promising industrial applications. The traditional vision is that a system will predict that a specific machine will fail within a certain number of days, giving the plant time to schedule repairs. DeBois is skeptical that this type of prediction is realistic for many forging operations. A plant may not have enough relevant historical failures for a particular press, hammer, furnace or automation system to develop a trustworthy prediction. Major equipment failures may be too infrequent, too varied or too poorly documented for an AI model to confidently forecast an exact failure date. A more practical opportunity is smart condition-based maintenance.

32 FIA MAGAZINE | AUGUST 2026