Niko switches to AI-driven quality control with Robovision

Every day, large quantities of faceplates and other finishing materials roll off the assembly line at Niko’s facility in Sint-Niklaas. These faceplates have to fit switches and sockets, and they have to look perfect. For quality control at Niko, there was still a lot of room to further optimize this process.

Towards an automated sorting process

In the past, quality control was partly automatic and partly manual. As a result, a significant part of good faceplates were wrongly rejected, while imperfect ones were approved. As a result, costs were higher than anticipated: ‘good’ faceplates were not sold or recycled, and a portion of the ‘bad’ faceplates ended up in stores anyway. Niko needed to find a way to automate quality control to reduce these errors and boost cost efficiency.

Together with delaware, Robovision offered an AI platform which classifies defects without human intervention. “In the past, developers had to catalog defects manually, but now with deep learning and machine learning, the AI platform writes the algorithm itself,” explains Jonathan Berte, CEO of Robovision. 

Spotting the blind spot

“In the case of Niko, we used a few hundred images of perfect and defective faceplates to train the algorithm. On the unacceptable faceplates, we indicated the defects down to the single-pixel level. Thanks to AI and computer vision, we could spot defects that weren’t even visible to the naked eye. With a high speed camera, the algorithm immediately and correctly identified every defective faceplate,” explains Sven Arnauts, manager of delaware.ai.

“The AI platform wrote its own algorithm in only a few hours. The big advantages are accuracy, cost efficiency and time savings, but you also don’t need any programming knowledge to create an algorithm. Even more, this technology runs in the Microsoft Azure cloud, so it doesn’t require an expensive server farm,” according to Sven.

“With the help of Robovision, Niko was able to decrease the number of ‘good’ products that were wrongly taken off the conveyor belt by 72%. The number of ‘bad’ products that wrongfully remained on the conveyor belt was cut by 33%,” Sven reveals.

AI’s full potential

Deep learning is not simply useful in automatically mapping out production deviations. Another advantage is that you don’t have to train the AI system all over again when new product variants arrive, such as when white faceplates are swapped with those in other colors.

However, even if AI can detect every defective faceplate, a robotic arm or a pusher is needed to remove the faulty plates from the belt. “Keep in mind that only by connecting AI to your business processes and machinery can you realize AI’s full potential,” says Jonathan. “AI helps you detect defects, but to create more efficiency, you also have to ask the question, ‘why is there a defect?’ To answer it, it’s necessary to integrate AI into your enterprise resource planning (ERP) solution and production line machines. Integrating the results of visual quality control into end-to-end supply chain processes (such as purchasing and sales), for example, can be done by connecting Robovision's solution to a company's quality measurement system.”

Is your organization ready to embrace smart manufacturing? Our team at delaware will help you boost your efficiency gains and competitive advantages. Reach out!