Computer Vision AI to Automate Defect Detection
Unilin Group’s partnership with Robovision led to a significant increase in yield through automating defect detection using computer vision AI for its laminate flooring production line.
Unilin’s Manufacturing Hurdles: Balancing Speed and Quality
Unilin Group is one of the world’s largest flooring manufacturers, known for its popular Quick-Step laminate brand. Facing three big production challenges, the company set out to transform its manufacturing process to boost operational efficiency.
Fast Assembly Line Speeds
High-speed production lines challenge human visual inspection capabilities, making real-time defect detection nearly impossible. Laminate plates, for example, are produced at a rate of 100 per minute—making it clear that manual defect detection is outside the reach of human capability.
Manual quality checks require time-consuming offline pack reviews. Unilin realized that replacing these manual quality checks with a fully automated vision AI solution could significantly reduce inefficiencies.
Shortage of Trained Manpower
As everyone knows, manual inspection is labor-intensive and time-consuming. Implementing AI automation means that operators have more time to focus on other production tasks. It also allows for early intervention, such as analyzing defect patterns and identifying root causes, further optimizing production workflows.
Manual quality inspection has reached the edge of human capabilities; that’s why we use vision AI.
Klaus Lozie, Digital Operations Manager at Unilin Group
Detecting Unpredictable Surfaces
Detecting defects in laminate meant to imitate distinct properties like organic wood grains is a complex undertaking—much more challenging and subtle than detecting people in a picture, for instance. Traditional computer vision cannot tell the difference between a defect and a print element with an unpredictable pattern.
The automated quality control system developed by the Unilin-Robovision partnership uses deep learning to detect even microscopic flaws in Unilin’s high-speed production environment.
Robovision in Action
An in-line camera scans critical zones of the laminate board, and 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; automated inline quality control ensures consistent accuracy. Operators can also adjust production factors to avoid producing more boards with the same defect to prevent recurring issues.
How Collaboration Leads to a Solution
To integrate Robovision’s AI software with Unilin Group’s specific infrastructure, the two entities collaborated to select an optimized architecture for building AI models using NVIDIA GPU technology. A highly accurate classification model was developed and rigorously tested on real production samples and challenging test sets.
Once the model was ready, Unilin’s experts were trained to use the Robovision AI platform. They learned to deploy, manage, and refine their AI models independently without requiring AI expertise or a dedicated data science team. This included learning how to retrain the model on new laminate styles, such as new patterns or colors.
Thanks to Robovision AI, monitoring defects of every single laminate plate is now feasible in such a complex and fast-moving production process.
Dries Van Poucke, Process Engineer at Unilin Group
Overview: How Vision AI Boosts Yield
- Automates complex visual tasks: Detects subtle defects in a high-speed production environment.
- Continuous monitoring: Analyzes every laminate plate with precision and consistency.
- Accurate surface defect detection: Deep-learning models autonomously learn all defects on different laminate types.
- Increases efficiency: Helps operators focus on other production tasks.
- Easy roll-out: Provides access to an easy-to-use platform with no AI knowledge required.
- Highly scalable: Ability to intuitively scale the platform across product types and production lines.
Want to learn more?