How an algorithm helps BekaertDeslee reduce waste

Creating the perfect environment for blissful sleep is what Waregem-based company BekaertDeslee is all about. So what could possibly keep the world’s leading manufacturer in mattress fabrics up at night? The answer: the huge amounts of production waste that come standard with the process of weaving and knitting complex textiles. One daring engineer, an ingenious algorithm developed using Azure Machine Learning, and a help delaware did away with the bad dreams.

“Producing textile is a tricky business,” begins Rik Holvoet, who has been CIO at BekaertDeslee for over 9 years. “The quality of the yarn, climatic conditions, humidity… all these factors can impact both the performance of the production machines and the quality of the final product. Add to that the challenging designs our creative people come up with, and it’s not hard to see why fabric manufacturing is notorious for its high waste production.”

The operators’ intuition

At BekaertDeslee, the quantity of waste produced is not neglectable: it translates into many soccer fields of waste per day. “We knew we had to do something, so we started looking for the factors with the biggest impact,” Rik continues. “Surprisingly, much of that came down to the instincts of our operators: the personnel with the most reliable intuition regarding when a machine was likely to break down, or when the parameters weren’t exactly right had the best chance of preventing a bad batch. Sadly, that’s not something you can learn in a few weeks: it comes with years of experience. But maybe artificial intelligence could help us out.”

A goldmine of shopfloor data

For over 8 years now, BekaertDeslee has been collecting shopfloor data with SAP and VisionBMS software and carefully storing it for the day its insights would be needed. Now, with the waste-reduction project, that day had finally come. “Storing all that production data is costly,” explains Rik. “But we knew it would pay off in the end. This project is proof that we were right.”

Armed with a wealth of information, a daring engineer at BekaertDeslee decided to try his hand at Azure Machine Learning to develop an algorithm that could predict when a machine was likely to break down. “Without any prior knowledge of the platform, he managed to get reliable results and tell operators which machines they should check to prevent mistakes from happening,” says Rik. “That’s when we knew we were onto something.”

40% waste reduction in Turkey

Rik and his team quickly decided to launch a limited pilot project at BekaertDeslee’s plant in Turkey that involved just a few machines. Later, the goal is to expand the project to the company’s 20 production sites across the globe. “To ensure that the technology is robust and scalable, we called upon our trusted partner, delaware,” Rik continues. “We’ve been working with them on numerous projects, often within the DEL20 innovation program, so they know our company inside-out. Their expertise was essential in exporting the production data to the cloud, integrating it with our ERP and using it to improve the prediction model over time.”

With the help of delaware, Rik and his team implemented the algorithm in Turkey. Although production performance is this year heavily impacted by the Covid19 pandemic, a waste reduction up to 40% turned out to be realistic. “Needless to say, the ROI of this project was extremely high,” says Rik. “To me, it’s a prime example of the long-term value of storing data, and of sharing insights and innovation.”

Discover how machine-learning technologies can be applied in real business situations to add measurable value.