PMMI ProSource – Start Your Search
Check out our packaging and processing solutions finder, PMMI ProSource.

Unilever Optimizes Hellmann’s Mayo Fill Levels with AI-based Predictive Analytics

Live from Automation Fair: Unilever addressed mayonnaise fill level inconsistencies using Rockwell Automation's FactoryTalk LogixAI, enhancing efficiency and reducing overfill giveaway through predictive analytics.

Unilever's Jean Ramos and Denis Catro de Almeida turned to Rockwell Automation's FactoryTalk LogixAI, a solution designed to predict fill amounts using what could be called, 'a soft sensor approach
Unilever's Jean Ramos and Denis Catro de Almeida turned to Rockwell Automation's FactoryTalk LogixAI, a solution designed to predict fill amounts using what could be called, "a soft sensor approach

For food producing brands, maintaining precision in fill levels is crucial for both product quality and cost efficiency. On its 320 to 350 jar/min Hellmann’s mayonnaise packaging line, Unilever’s Pouso Alegre food and nutrition facility outside of Sao Paulo, Brazil, faced challenges with its traditional PID (proportional integral derivative) loop system, which relied on feedback from downstream checkweighers to adjust fill levels. This reactive approach sometimes led to overfilling, increasing material costs and reducing overall efficiency, according to Jean Ramos, digital factory coordinator, and Denis Castro de Almeida, digital factory coordinator LATAM, both of Unilever.

The ideal weight was 500 g, but the scrap window was already narrow—499 g for underfills, and 508 g for overfill scrap. How could Unilever tighten the standard deviation and move the average fill closer to the 500-g ideal weight, without losing scrap to underfill? Ideal weight was 500 g, but 499-g underfills were scrap, as were overfills over 508 g. Before and after bell curves show a tighter grouping of fills much closer to the ideal 500-g jar without going under, greatly reducing giveaway and scrap.Ideal weight was 500 g, but 499-g underfills were scrap, as were overfills over 508 g. Before and after bell curves show a tighter grouping of fills much closer to the ideal 500-g jar without going under, greatly reducing giveaway and scrap.

At Automation Fair, Ramos and Catro de Almeida explained that they turned to Rockwell’s FactoryTalk LogixAI Perfect Fill, a new solution designed to predict fill amounts using what could be called, “a soft sensor approach, so think of it as software as a sensor,” said Richard Resseguie, product manager for Rockwell’s LogixAI. “The whole theory behind this is what if we can develop a soft sensor that's going to predict what the fill amount is going to be in the jar, before you actually go and fill it, based on the process variables.”

The implementation of LogixAI involved integrating data from various stages of production, including upstream filtering and mixing processes as well as the filling itself. This data was used to train a predictive model that could anticipate the expected dosed weight, to a high degree of precision, based on those upstream variables.

“We're doing this directly on the edge right next to the equipment. That’s how we're gathering the data, how we're training the model, and then how we're adjusting to improve it,” Resseguie added.

Connected Workforce Report
Discover how connected workforce technologies and automation can bridge the skills gap in our latest report. Explore actionable insights and innovative solutions.
Read More
Connected Workforce Report
New e-book on Flexible Packaging
In this e-book, you’ll learn key considerations for vertical and horizontal f/f/s and how to choose between premade bags and an f/f/s system. Plus, discover the pitfalls to avoid on bagging machinery projects.
Read More
New e-book on Flexible Packaging