AI at Unilin: assess what is there and make improvements
Six months ago a global AI hype took the world by storm. Unilin was in no hurry to climb the bandwagon because we have been investigating the potential benefits of AI for all our processes for several years now. Meanwhile some AI models have been integrated into the production process while others are currently being trained and tested. Still others were found to be too unreliable after a test phase. The key question is: when does AI provide us with added value?
AI has been the talk of the town for months now but what is artificial intelligence (AI) really? Tom De Smet, product owner within the Digital Operations team of Unilin Flooring, describes it as follows: “AI is the capacity of a machine to imitate human behaviour. You could see AI as a child that learns by example but, contrary to a child, AI forgets nothing. Constant training makes it increasingly intelligent.”
At Unilin we’ve been looking into AI for some time now, in line with our purpose. This is because AI can help us raise the bar and produce better products that will enrich the lives of our customers.
More complex processes with the help of AI
Product concepts and designs are becoming ever more complex. Just think of our ambition to manufacture floors and boards that can’t be told apart from genuine wood. This requires a great level of detail, resulting in increasingly intricate production processes based on thousands of settings. The more complex the product, the more sources of variation arise, leading to greater risks of machine downtime and production loss.
The processes are becoming too complicated for our operators, which is why several departments have been asking themselves how AI can support our operators and engineers. Tom: “The days when the same product came off the production line for hours on end are long gone. We train our AI models with data to teach them how to handle variation in designs and colours. The more data, the greater the accuracy of your AI model. Today we also have the necessary infrastructure to store such data volumes and that was not yet the case ten years ago.”
In 2019 the first AI stone was laid when Unilin Flooring went in search of the right external partner to help us implement AI in our production processes. This led to the launch of the first case in early 2020 and some of those initial projects have meanwhile been integrated into production. This makes Unilin a pioneer in the industry.
The approach implemented by Unilin Flooring is a fine example of how Unilin handles the development of an AI model, says Stijn Lioen, Domain Architect Data & Analytics at Unilin. “We started from an existing AI model by Google to detect errors but that model wasn’t good enough at handling variations, so we adapted it. That is typical Unilin. We take what exists on the market and we make it better. This way we stay ahead of the competition and are able to develop better and more innovative products.”
Both Tom and Stijn are convinced that the possibilities are limitless. “I don’t think there are any limits”, Stijn laughs. “For instance, with AI we can make decors from wood that doesn’t exist. The trials are ongoing but right now they’re still in the early stages.” Tom: “We can take these designs really far but at the same time we verify whether the new product would fit our vision of ‘enduring beauty’ and ‘true to nature’.”
Tom sees plenty of potential to interlink the data of the various departments. “At the moment we are capturing a lot of data we’re not using yet. For example, interlinking process parameters across and within the divisions would greatly accelerate things.”
Through the hype
A major concern with AI is the potential loss of jobs. Tom qualifies: “I think it would be more accurate to say that AI will change the job content. Our industry faces huge shortages in terms of technically skilled personnel. We always have vacant positions. AI could take over tasks from roles for which it is already difficult to find new staff. AI can help us bridge the gap between the more complex production processes and less skilled personnel.”
One thing is for certain: the role of AI will continue to grow, precisely because we are starting to embrace it. “That is one of the strengths of Unilin’s pragmatic entrepreneurship: investing in knowledge to see through the hype. We are not only given the time to thoroughly immerse ourselves in the subject matter but also the freedom to make mistakes so we can experience what doesn’t work yet. If we are convinced a specific avenue can produce results then we continue to invest in it. Not every company would devote so much time and resources to this, but if you’re going to move forward and innovate they are absolutely essential.”
Two examples of successful AI implementations at Unilin
1. AI Finders
We strive to create laminate that looks and feels like actual wood. To achieve this, we use cameras that perfectly position the wood texture and wood pattern in relation to each other. In conjunction with an external partner, we also introduced AI into this process. The AI model is based on deep learning and had to meet strict requirements. Tom: “In 99% of cases our classic computer vision gives us a good result but that final percent eludes us. This results in downtime and outages. So AI has to produce better results or the operator won’t see the added value of working with AI. That is why we strive for 99.9% or even 100%, results that would be outright impossible with any other technology. In addition, it has to learn to deal with different sources of variation, it must be pixel-perfect and applicable on all machines.”
The next step consists of expanding the solution to other production lines and new problems. From 1 million predictions per month we have now set our sights on more than 3 million predictions per month by the end of the year.
Several models have meanwhile demonstrated their added value and this leads to greater acceptance on the production floor. Stijn Lioen: “Argus is a quality detection system on the line that manufactures the Quick-Step Capture collection of Unilin Flooring. The deep learning AI model behind the system is also able to not detect the intentional imperfections in the design (which also occur in real wood) as an error, which helps the operators to check every single board on the line. Before the introduction spot checks were carried out every 20 boards, which led to more production loss. Now Argus traces the smallest defect thanks to this wonderful innovation. Since March 2023 Argus has been deployed on all Capture colours and in that short period the operators have already indicated that Argus has become indispensable.”