Predictive maintenance investments are still seen as low-hanging fruit with a high return on investment (ROI). At the recent Connected Workforce Conference in 2025, a keynote speaker surveyed the audience about which use case would deliver the biggest ROI. Not surprisingly, the results showed that IoT for predictive maintenance ranked first at 55%, while AI/ML demand forecasting ranked second at 32%.

This anecdotal evidence is confirmed by Jorge Izquierdo, PMMI’s VP of Market Development, in a recent FOOD ENGINEERING article on OEE. “Predictive maintenance remains a valuable, evolving aftermarket feature,” Izquierdo says. “In fact, 35% of end users surveyed indicated they plan to increase spending on predictive maintenance.” Izquierdo is referring to the 2025 State of the Industry report from PMMI, The Association for Packaging and Processing Technologies.

However, the heat is on for manufacturers after years of investment in automation and digital tools by Hershey, Heinz, and even smaller brands to increase throughput and demonstrate innovation now. Along with traditional automation and sensing, these new capital investments include advanced machine learning and digital twin services from original equipment manufacturers (OEMs) and third-party vendors.

Dwayne Negrón, digital twin capability manager at Kalypso, a Rockwell business, spoke about packaging bottlenecks at Rockwell Automation’s recent Automation Fair. Negrón said a client needed a virtual digital twin tool to reconfigure a case and tray packing line and avoid shutting down the line.

According to Negrón, the client had already spent “roughly six months of troubleshooting before bringing us online.” The solution was the Kalyspo Emulate3D platform, including modeling, virtual PLC code and HMI screens. “We are now able to provide virtual line time and do these tests whenever we want without affecting the physical system anymore,” Negrón added. “Using a digital twin, you can 10X the number of tests you do and never touch a physical line.”

Explore More Packaging

Digital twin technology is a solution for many industries, but it is also a large investment. Smaller digital solutions for packaging lines are also gaining traction. SmartSights’ ABLE data collection technology enables plants to combine PLC programs and HMI configuration files to identify key pieces of information for manufacturing execution systems (MES) and overall equipment effectiveness (OEE) applications.

“It’s difficult to provide an accurate root cause when only looking at the bottleneck asset, so this necessitates the need for advanced data collection,” said John Oskin, senior vice president at SmartSights in the FOOD ENGINEERING feature on OEE standardization. Data collected from SmartSights allows maintenance teams to replay and validate data in a digital twin using real control signals from a packaging line.

Packaging machine builders are ready to offer advanced data monitoring solutions and services to customers. “The key question for many manufacturers, however, is no longer whether to monitor packaging equipment — but how to turn monitoring into measurable performance gains,” says Björn Voges, global marketing manager at GREIF-VELOX.

The European company is seeing a shift that is driving greater integration of condition monitoring, predictive analytics and service expertise into OEM offerings. “Retrofit-capable solutions are playing a central role, particularly in European food plants where existing lines still offer significant optimization potential,” Voges says.

With the help of agnostic industrial data protocols like OPC and MQTT, packaging lines are moving away from islands of automation to being connected to other systems within the plant. Chobani’s Idaho Falls plant made a splash in 2015 when the yogurt producer added Inductive Automation’s Ignition SCADA platform and began to scale data monitoring across its plant. Adding data monitoring to packaging lines was the next step in Chobani’s OEE efforts.

“We’ve been focusing on simplistic things like data,” says Hugh Roddy, Chobani’s VP global engineering and project management, at the 2025 Ignition Community Conference. “We were able to use our downtime analysis. Forget that you have 48 different downtime events. Focus on the top five downtime events and resolve them. Now you've got the next five, and the OEE just went up and up.”

Data standardization efforts are accelerating packaging line optimization projects and easy OEE wins. Recent data standardization discussions center on a Unified Namespace (UNS) approach, which focuses on an organized, centralized data broker that provides context for your entire business.

“We see UNS and other approaches on how to contextualize and normalize data coming from equipment,” says Matt Wise, CEO of E Tech Group. “Companies have IT people that understand UNS and have data in one spot and in a format that can be analyzed.”

Machine Learning and New KPIs in Packaging

Machine learning (ML) has been the hero in manufacturing for the last 10 years. While AI receives all the headlines, successful predictive maintenance services and solutions have driven advances in machine learning for motors and drives.

Mark Bertrand, director, industry solutions at SmartSights, recently discussed how regression modeling and machine learning are evaluating real-world packaging line bottlenecks in new ways. Bertrand describes modeling as prescriptive analytics, the analysis of a line's characteristics. “The filler on the packaging line is often seen as your bottleneck, but we've run models to find out that it’s not true,” Bertrand says.

In a 2025 Control System Integrators Association (CSIA) webinar, Bertand introduces the term feature importance, a generic concept in ML models. Feature importance can be based on any manufacturing metric, such as meantime between failures (MTBF), or on a machine center value.

“For a filling model, we created our own KPI, and this takes the availability times average rate, and we call this the effective rate,” Bertrand says. “It's an engineering unit, so it's really units per minute, and this is our primary KPI because that's the true throughput capability of that machine center.”

For a real-world application, SmartSights used its ABLE technology to conduct a root cause analysis of a packaging line that included a bundler, a wrapping unit and a tray packer. Ranked by machine center, ABLE identified the highest impact machines in terms of root causes.

“The more complex the line, accumulation, parallelism and the root cause can get a little abstract,” Bertrand says. “But the root cause analytics were saying that the tray packer was our biggest cause of downtime for the overall line, but the model actually ranked the bundler.”

“Both of the algorithms were correct, but what was misleading was that the root cause is saying operators should be focusing on this tray packer. However, the focus should be on the bundler,” Bertrand adds. ML modeling allowed the packer to increase the overspeed capacity and increase speed (rate) in both machines, solving the machine center bottleneck.

“We noted after retuning that the average accumulation upstream of the bundler was considerably lower and the number of stops cascading through the line was also reduced, increasing the overall throughput of the line,” Bertrand says. “So that was a positive win using the data from the ML versus the data we were using for more traditional analytics.”