From Raw Data to Reliable Decisions: How Yara’s AnomaliSense™ Puts Predictive Maintenance in the Hands of Technicians

In the high-stakes world of process industry operations, where downtime can mean millions in lost production, predictive maintenance offers an exciting opportunity to transform performance, optimise resources, and boost reliability. When done right, it can turn untapped data into a competitive advantage. 

At Yara, one of the world’s largest nitrogen-based fertilizer producers, this opportunity sparked the creation of AnomaliSense™, an in-house platform focused on human usability and integration into existing workflows. Ahead of their presentation at Asset Performance 2025, we sat down with Perry Jaspers and Madava Dilshan Vithanage to discuss why they built it, how it works, and why the journey started not in a boardroom, but out in the field alongside technicians.

Designing predictive tools for practical impact

Perry, Technical Process Owner Electricity & Automation, has been with Yara for two decades, starting in project organization and now focusing on operational excellence. In his current global role, he oversees Yara’s electrical automation processes, from instrumentation to site-level predictive analysis. He knows the challenges driving the shift toward predictive maintenance first-hand.

“Yara has a big production footprint in Europe, where gas is expensive,” he explains. “We have to be more efficient than competitors in regions with cheaper feedstock. With strong traditional maintenance in place, the logical progression for us was to embrace predictive techniques. But our definition is very specific. It starts when a continuous data stream triggers the next action, not when someone just ticks boxes on a field checklist.”

Most off-the-shelf platforms offer a vast range of features, but for Yara the real value lay in a targeted approach. “It’s like Excel,” Perry says. “Most people only use 5 % of the capabilities. The predictive tools on the market were overkill for what we needed. Our platform needed to be streamlined, adaptable and delivering exactly the insights our teams need without unnecessary complexity.”

A sharp focus: low cost, high scalability, seamless integration

That’s where Madava, with a background in computer science and electronics, entered the picture. As product manager in Yara’s Digital Core Solutions team, his mission was to take Perry’s operational challenges and translate them into a tool the Yara plants could actually use.

“Cost and scalability were the biggest hurdles,” says Madava. “Most vendor solutions are priced in a way that makes sense for a handful of critical assets, but not for scaling across thousands. We needed the opposite, a low-cost system that could cover a huge asset base without requiring a team of ten people just to operate it.”

The team had a head start thanks to Yara’s Digital Production Platform, a central data lake streaming process control data from all sites at one-second granularity. This gave them the raw material for advanced analytics without the need for expensive data-gathering projects.

From there, they focused on two design principles: integration over reinvention and human-centred usability. AnomaliSense™ connects to existing third-party tools used within Yara, such as SAP, System 1 and Control Valve App, pulling their specialised analytics into a unified front end. At the same time, the output is designed to be clear and actionable for any maintenance technician, without the need for extensive training or vendor certifications. This combination ensures that the platform works seamlessly with established systems while remaining practical and accessible for the people who use it every day.

Adding context to alerts

Most condition monitoring still works on simple rules: a temperature exceeds its limit, an alert is raised. But in large plants, thousands of such alerts can overwhelm teams, and without context, they often go ignored. AnomaliSense™ addresses this by using a mix of statistical, machine learning, and deep learning models that assess asset behaviour in context. Seasonal variations, process conditions, and correlated sensor data are all considered before deciding whether an alert is truly urgent. 

A dynamic prioritisation algorithm ranks the alerts and thus ensures that the most critical issues rise to the top, while nuisance alarms fade into the background. “At one plant, our Integrity Operating Window alerts revealed that a key sensor had been generating warnings for months,” says Madava. “It turned out to be a sensor fault, not a process fault – but if it had been real, the cost impact would have been enormous.”

Making predictive maintenance part of the day job

Creating AnomaliSense™ wasn’t just about software architecture, it was about trust and adoption. The backbone of maintenance at Yara, is SAP, so the platform was built to integrate directly into it. Predictive tasks appear alongside preventive work orders, following the same planning, approval, and scheduling steps. This ensured the tool felt like a natural extension of daily work rather than an extra system to manage. 

“If you push a tool from the top down, you get resistance,” says Perry. “We started with specific problems the plants couldn’t solve with traditional methods. When you give them something that actually helps, you get buy-in. Now we have eight sites using AnomaliSense™, and it’s spreading fast.”

Yara’s technicians saw an immediate difference. Instead of wrestling with multiple interfaces or deciphering complex analytics dashboards, they received clear, prioritised information in a familiar format. Technicians called it ‘a tool built for us, not for someone in an office.’ The clarity of alerts and the ability to act without hunting through multiple systems meant less frustration, quicker decisions, and a greater sense of ownership over asset health.

Since its rollout, AnomaliSense™ has allowed sites to act on critical alerts more quickly, while significantly reducing time spent chasing false alarms. In several instances, early detection enabled maintenance teams to plan interventions well in advance, avoiding unexpected downtime and keeping production on track.

What to expect in Antwerp

The presentation at Asset Performance 2025 will dive deeper into AnomaliSense’s architecture, use cases, and integration strategy.

“If you can learn one thing from our case,” says Perry, “it’s to start from the field. Pick a problem your traditional methods can’t solve, and use predictive analytics to tackle it. That’s how you create value and buy-in from day one.”

Madava agrees, but adds a note of caution: “Predictive maintenance is like an insurance policy. You may go months or years without a major catch – but when it does detect a looming failure, it can save you from disaster. Manage expectations, be patient, and scale widely enough to make the returns visible.”

For technical managers weighing their own digital transformation strategies, Yara’s story is both a reality check and an inspiration. Predictive maintenance doesn’t have to be a costly, overcomplicated endeavour. With the right focus, the right data foundation, and above all a commitment to the people who will use it, it can become a natural, trusted part of daily maintenance life.For any plant seeking to turn its data into decisive action, AnomaliSense™ is proof that predictive maintenance can be practical, scalable, and embraced by the very people who keep assets running every day.

Yara AnomaliSense™ – Delivering Human-Centred, Actionable Insights
📅 November 4, 9:35 – 10:10, Asset Performance 2025, Antwerp