Automating Defect Detection with Computer Vision AI

Unilin Group partnered with Robovision to develop a computer vision AI-based defect detection system for their laminate manufacturing process.

Defect detection at high production speed

Real-time detection of rare and hard-to-see defects is not only a challenge, but it also creates an opportunity for manufacturers to raise the bar on their quality standards, increase efficiency for operators and improve production yields. 

Unilin Group is one of the world’s largest flooring manufacturers, mostly known for its successful brand Quick-Step. They were looking for a solution partner to build an AI-powered in-line quality control system that would help their operators detect defects in their laminate flooring.  

Unilin Group partnered with Robovision to build a Computer Vision AI-powered solution that automates visual inspection, ensuring optimal product quality and higher production output. 

Hard-to-find employees for a tough job

Unilin’s challenge was threefold. First, the high assembly line speed (100 meters/minute) makes it difficult for even experienced operators to accurately spot all possible flaws. Second, the shortage of trained operators or domain experts is a huge constraint for manual quality inspection. Third, as their laminate flooring is almost the same as real wood, it is not easy to distinguish a defect from a natural element of the print.

The quest for continuous quality Improvement

Embracing innovation, Unilin Group constantly strives to deliver the highest quality through emerging technologies. As the laminate plates pass by at high speed, it is very hard for operators to see or detect the small defects inline. In the event of an error, the operators need to check different laminate packs offline to ensure the quality of end products. To avoid double work and increase efficiency, continuous defect monitoring was necessary. As manual quality inspection has reached the edge of human capabilities and qualified operators are hard to find, full automation was considered. 

Focus on early intervention

The search for a new solution began when Unilin Group realized that AI would help reduce defective end products and increase efficiency for staff members. Unilin Group approached Robovision to develop a scalable computer vision AI solution with the easy-to-use Robovision AI platform to automatically detect defects across different laminate types and colors. 

Higher yields and a competitive advantage

By removing time-consuming manual inspection, domain experts are able to focus on early intervention: analyzing the defect and taking timely actions in the production process. The automated inline quality control process is streamlined and more accurate than ever, resulting in higher production yields and helping Unilin achieve a competitive edge. 

Solving the problem at production scale

AI-powered defect detection is the future of visual inspection. By combining AI-enabled systems with cameras, manufacturers can deliver high quality with reduced time delay and defect costs. 

Together with the Unilin team, Robovision built an automated quality control system that can detect even microscopic flaws in Unilin’s high-speed production environment. An in-line camera scans critical zones of the laminate board, then the AI model processes the data and provides visual insights. If a defect is detected, the operator is immediately alerted to remove the defective laminate plate from production. Operators can now focus on other tasks while AI does its work. They can also adjust production factors to avoid producing more boards with the same defect. 

A scalable solution

Within the Robovision AI Platform (RVAI), the Robovision team developed an AI model that is scalable across different product types. The Unilin team can easily retrain the model on different laminate types and colors, with easy deployment across production lines and factories. 

Deep learning: a game changer

Detecting defects on laminate is much more challenging and subtle than for example detecting people in a picture.  Because laminate has almost the same natural characteristics as real wood, it is difficult to tell if there is a defect or if it is just an element of the print. Traditional computer vision would not be able to handle these subtle changes of the laminate dataset. That’s why we used a deep learning AI model that can detect defects on different types and colors of laminate. 

A powerful platform for AI-driven companies

The Robovision AI platform was provided as part of the total solution. Robovision first helped the Unilin Group select a carefully designed architecture to build the AI models leveraging NVIDIA GPU processing power, within the platform. Then, using an iterative approach, we were able to develop a highly accurate AI classification model that was tested on both production samples and difficult test sets. We then gave Unilin’s experts training on the Robovision AI platform. 

A future-proof solution

The Unilin team can use the RVAI platform on-site. Without any AI knowledge or having a data science team, they can deploy and manage the lifecycle of their AI models flexibly and autonomously. They can also retrain the model on new types and new colors of laminate by themselves. 

Higher production output

Thanks to the Robovision AI platform, Robovision and Unilin Group were able to join forces. Robovision’s AI expertise was combined with Unilin’s product insights and problem understanding to implement an AI-powered in-line quality control system. With this AI-based defect detection and a streamlined visual inspection process, Unilin Group achieved a higher production output. 

• Automate complex visual tasks: Detect subtle defects in a high-speed production environment.
• Continuous monitoring: Spot specific defects on every single laminate plate.
• Increased efficiency: Help operators focus on other production tasks.
• Accurate defect detection: Deep-learning models autonomously learn all defects on different laminate types.
• Easy to roll out AI models: Access to an easy-to-use platform with no AI knowledge required.
• A highly scalable solution: Scaling the solution across product types and production lines.

Ready for AI-powered manufacturing?