Blake Moret, CEO, Rockwell Automation, opened the 2025 Automation Fair by stressing that manufacturers are using automation to move toward autonomy, machine learning and AI strategies. In essence, doing more with less.

Food manufacturers are also embracing doing more with less when it comes to predictive maintenance. Companies are investing in digital twin and cloud monitoring platforms to scale predictive maintenance across plants, reduce preventive workloads, deliver advanced visibility and repurpose workers. This article will examine case studies, technologies and approaches for launching these predictive maintenance programs.

Predictive Maintenance Finds Success in the Cloud

Companies have identified digital twin and cloud solutions for predictive maintenance as a way to achieve greater autonomy and do more with less.

"Starting with a focused pilot limited to solving known issues on one or two high-impact assets allows teams to build a repeatable playbook with confidence before scaling," says Jim Toman, MES functional consultant at Grantek.

The popularity of larger investments in predictive strategies also resides in the data. "The data requirements are comparatively straightforward with predictive maintenance: vibration, temperature, current draw — often from sensors already in place," says David Ariens, founder at the IT/OT Insider. "The physics are well modeled, the failure modes documented, and the business case is easy to articulate. That combination makes it fertile ground for quick wins."

In addition, cloud platforms and monitoring solutions are getting easier to integrate. "Companies are pivoting away from isolated systems like Computer Maintenance Management System (CMMS) and are instead focusing on connected systems that work together cohesively," says Michael DeMaria, director, product management at Fluke. "These integrated systems have the ability to do much more than act as an alert system. They can help to coordinate maintenance, energy and inventory in one single loop."

Grantek, a full system integrator of automation and system software solutions, also helps clients standardize their asset data architecture, often using a combination of native OPC-UA connectivity on newer equipment and edge retrofits on legacy assets, so that the same failure mode looks and behaves consistently across plants.

(Source: Cisco)

Data Modeling in Action

At the AVEVA 2024 Conference, Carlos Paredes, controls engineering manager, Amcor Flexibles, discussed how the company rolled out an anomaly detection initiative across its plants for blow and injection molding applications.

The company uses AVEVA’s MES platform, CONNECT data services and the Advanced Analytics tool to understand upstream anomalies for its 200 blow and injection molding assets across multiple plants. "If you're going to leverage advanced analytics, I’ll encourage you to start small, understand the technology, see how processes are correlated, and then think about building more trends."

For the rollout, Paredes’ team tested its molding machines for anomalies at its first plant in Ames, Iowa. However, real-time advanced analytics were not connected to proper alerting processes. "When developing this process, we didn't have the right notifications. We didn't really know how we were going to interact with the operations," Paredes said in his keynote presentation.

Once the solutions started predicting events and found anomalies in Iowa, operators and maintenance teams were able to plan accordingly. With advanced analytics, the team identified that the dryer was constantly dropping temperature, which would eventually lead to unscheduled downtime.

"Once we had a stable process in our Orlando facility, the system generated an anomaly score and this metric was merged with our MES platform," Paredes said. "We could see the amount of scrap generated during that time and identify the main cause."

Currently, Amcor managers are reviewing anomalies weekly with the plants and SMEs. "What are the anomalies that we can take care of, or can we increase the cushion (before fixing or replacing)," Paredes said. "And, we start feeding the model to make it more operational and receive feedback from operations."

So, what were the results?

Amcor deployed the cloud-based system across multiple plants and realized 2% of potential reduction of unscheduled downtime across its plants in the beginning of the program.

Maintenance Training on the Rise?

System integrators can be essential in implementing new tools for predictive maintenance initiatives. "Beyond the technology, we invest heavily in structured handoff programs, including hands-on training, documented standard operating procedures (SOPs), and embedded support during early live operation, because deployment without adoption is just expensive shelfware," Toman says.

Mars recently piloted a predictive maintenance program for chocolate production across multiple plants using Datadog, a digital twin solution that runs on the Microsoft Azure IoT Edge platform. The Datadog solution can be used in cloud and hybrid environments and offers modeling and advanced analytics in operations.

Explore More Maintenance Strategies

Mars’ objective was to increase autonomous monitoring and use the digital twin platform to understand simple anomaly detection to begin modeling normal and abnormal conditions. Datadog’s Watchdog is the intelligence layer of the solution.

"The monitoring solution is able to understand patterns, so when an anomaly appears, operators receive alerts to understand if a device is not working appropriately or is not able to connect as expected," said Luiz Fraga, senior lead AIOPS and data production architecture, Mars, Inc, during his keynote presentation.

For the pilot, Mars trained the digital twin on simple up/down or on/off device behavior to understand the conditions for success. In addition, the solution provided predictive data to operators.

"When something wrong happens, the operations and development teams have enough information for troubleshooting," Fraga says. "They can use logs and other data to quickly fix the solution and send recommendations."

The digital twin solution enables operators to use composite monitors that combine up to 10 individual monitors. "Monitors can be of different alert types: simple alerts, multi-alerts or a combination of the two," Fraga said. "If you choose a multi-alert monitor, the user interface shows the monitor’s group-by clause and how many unique sources are currently reporting."

Fraga added that 18 months after the rollout, observability is needed as early as possible for a digital twin pilot program. "The signal team was engaged during the early stages, and without them, we probably would have failed in deploying at scale," Fraga says.

"It's become apparent that companies cannot simply rely on the system outputs," DeMaria says. Double-checking work and overall analysis of data outputs is a step that is often skipped but is vital to operational efficacies and one lesson that many organizations are learning."

The workforce component is a recurring theme when scaling predictive maintenance strategies. "Involving maintenance technicians is important from day one; when tools are built without the people who will use them, adoption fails regardless of how sophisticated the technology is," Toman says.