Competition within the manufacturing industry is growing, both in Europe and Asia. Manufacturing companies are therefore constantly looking for ways to work smarter, more efficiently, and more accurately. Yet AI is still in its infancy at many organizations. “If you want to remain relevant, you have to explore the possibilities,” says Stan Erens, COO of Meldon. Together with Blue Engineering, the company started an AI-MATTERS experiment, offered by Brainport Industries Coöperatie, to improve product quality through visual inspections. “We can apply the insights we gained in several areas.”
Meldon: constantly evolving
Meldon develops and manufactures plastic profiles for a wide range of applications. The family business works in a multidisciplinary, innovative way and collaborates closely with customers and partners. “Our production runs 24/7, with five shifts and 27 extrusion lines,” says Stan. “With 95 colleagues, we make our own PVC compound and granulate and develop and produce our own extrusion tooling.”
AI does not yet play a major role at Meldon. “We use robots and have a camera system in two production systems, but those systems are already fifteen years old. They are due for renewal. Through an AI-MATTERS service offered by Brainport Industries Coöperatie in collaboration with Blue Engineering, we worked with Blue Engineering to investigate which Vision technology is suitable for our processes and how we can upgrade our lines.”
The challenge: visible defects that are only discovered afterwards
The experiment focused on a plastic extrusion profile with high visibility requirements. During the production of these profiles, visible defects such as stains, scratches, and scrapes regularly occur. These lead to rejects, but are often only discovered later in the process. Operators operate multiple lines simultaneously and do not continuously monitor the product. With profiles ranging from six to twelve meters in length, it is difficult to trace a defect back to the moment it occurred.
Stan: “We don’t want to sort afterwards, but produce flawlessly right away. To achieve this, we need to make product quality measurable and recognize deviations during the process.” Machine Vision and AI offer a promising basis for this. They make it possible to collect real-time quality data and optimize process settings in a targeted manner.
A self-learning system with 2D line scan camera and Anomaly Detect
During the AI-MATTERS experiment, Blue Engineering investigated various recording methods and AI techniques. The tests showed that a 2D line scan camera is most suitable for continuous processes such as the extrusion of these profiles. This camera scans the product line by line and requires little lighting, which makes the system attractive from both a technical and cost perspective.
At the heart of quality control is the AI algorithm Anomaly Detect. This model learns exclusively on the basis of images of a good product. Anything that deviates from this is seen by the system as an anomaly. This fits perfectly with the variation in possible errors: shape, size, and structure vary greatly depending on the type of profile.
Stan: “Many systems only recognize errors that you define in advance. That takes a lot of time. With Anomaly Detect, it works exactly the opposite way: you teach the model what is good, and it automatically recognizes everything that does not belong. That is much faster and more reliable.” The system even recognizes subtle deviations, such as fine scratches. By adjusting the threshold value, engineers can easily fine-tune how sensitive the system responds.
The feasibility study shows that the combination of a 2D line scan camera and Anomaly Detect software reliably detects deviations in different types of profiles and distinguishes them from undamaged products. This is an important step toward direct quality control during production.
Results and next steps
The experiment provided Meldon with valuable insights. “We now know which technology works for us and how we can improve our processes with Vision and AI,” says Stan. In the coming period, Meldon will:
• collect and analyze data in a targeted manner
• improve process settings based on quality data
• investigate whether the existing camera systems on two lines can be replaced or upgraded
• deploy a new camera system for another application
“Our goal is to analyze data faster and better, so we can solve problems immediately instead of sorting them out afterwards.” The insights gained have now been shared in a presentation to all engineers. “This way, everyone can see what’s possible and where we can grow. The responses were very enthusiastic.”
AI-MATTERS as an accelerator
“Unknown makes unloved. Thanks to AI-MATTERS, our engineers now have a much better understanding of what AI and Vision can do for us,” says Stan. “That’s exactly why we participated: to build knowledge and discover what’s possible. AI-MATTERS helps companies get started with new technology in an accessible way.”
According to Stan, participation is definitely worthwhile: “These techniques are necessary to produce smarter, more efficiently, and more accurately. They help us remain relevant in a highly competitive market. That makes these kinds of projects very valuable.”