Smart algorithms, self-learning models, and factories that inspect themselves: it sounds like something from the future. But at EKB, they've come to know better. 'AI is a wonderful tool, but definitely not a magic wand,' warn experts Jasper Verhoef and Sebastien Negrijn. What do you as an industrial company really need to know before getting started with AI?
Machine Vision – the automatic inspection of products using camera images – has been a reliable tool in industrial production environments for decades. But where traditional systems often get stuck when faced with variation or deviations that are difficult to capture in rules, AI offers a solution. Still, there are also risks involved that, according to EKB, are often overlooked.
“Many companies think AI can do anything. That's simply not true.”
Of cats and cups
“You can tell the difference between a dog and a cat at a glance,” says Sebastien Negrijn, Manager Engineering at EKB. “But just try explaining that to a machine in terms of rules. That's where AI makes the difference: instead of setting up rules, you let the system learn the distinction itself.”
That's also precisely why AI is so powerful for quality control in dynamic production environments. Think of products that vary continuously – from cups with changing logos to plants in horticulture where you have to count how many flowers there are. “Traditional machine vision often can't handle this without high costs or lengthy reprogramming,” adds AI specialist Jasper Verhoef. “AI systems are much more flexible and faster to implement.”
Speed ≠ reliability
But there are downsides too. In high-speed environments – such as production lines where dozens of products pass by per second – AI can simply be too slow. On top of that, there's the risk that an AI model approves faulty products or vice versa.
“With a product that could have a potentially fatal defect – think of mold in food – you want 100% certainty. In that case, classic machine vision is often still the safest choice.”
The validation of AI models is another point of attention. “You train on historical data, but how can you be sure the model also works in practice?” Verhoef wonders aloud. That's why at EKB they always combine training and validation with strict test sets and weekly quality checks using deviating products.
What you as a manufacturer must know
According to Negrijn and Verhoef, there is one piece of advice that should top the list for companies wanting to get started with AI: know what you want to solve. Is speed most important? Is the product highly variable? Is 100% certainty crucial, or is a margin of error acceptable?
“AI is a tool. Not a miracle cure. Start with a clear
problem definition – and only then choose the technology.”
What does the future hold?
The experts expect AI systems to become increasingly powerful and user-friendly. “In a few years, you might simply be able to type in your acceptance and rejection criteria – and the model does the rest,” Verhoef predicts. “But until then, cleverly combining human insight, classic techniques and AI is the key to success.”
Want to experiment as a manufacturer with AI in quality control? Then bear in mind: it's not about implementing the latest hype – it's about solving the right problem. EKB knows exactly how that works.