)
)
)
)
)
)
)
)
How can we help you?
)
craftworks GmbH
Schottenfeldgasse 20/6a
1070 Vienna
Enhancing Scrap Detection to Improve
Quality and Throughput
navio VISION significantly improves scrap detection accuracy for manufactured bearings, reducing waste and optimizing production.
:quality(80))
Success Story
reading time: 2 min
Manufacturing
Predictive Quality
Visual Inspection
Anomaly Detection
50%
Reduction in scrap rate
94%
Accuracy in classifying bearings
82%
Precision in identifying defect types
Context
Our client, a leading manufacturer of industrial bearings, faced challenges in ensuring the consistent quality of produced parts. Bearings are critical components used in various machines, where even small defects can impact performance and lifespan. Accurate classification of defective parts is essential for maintaining high product standards and minimizing waste.
Challenge
The existing image-based scrap detection system was unreliable, leading to excessive rejection rates. Approximately 10% of produced bearings were classified as defective (NOK). However, manual checks revealed that 65% of these were actually high-quality (OK), and another 25% could be sold as 2nd grade quality — suitable for many applications. This resulted in unnecessary waste and lost revenue opportunities.
Assignment
We proposed to enhance the existing scrap detection system with navio VISION, an AI-based solution capable of accurately distinguishing between 1st and 2nd grade quality, while also identifying the type of defect.
Solution
navio VISION is trained using the same images as the existing system, focusing on identifying patterns in both OK and NOK parts. A second model is trained on human-classified NOK images to recognize specific defect types. The system is integrated into the production line, providing real-time classification and actionable insights.
The enhanced AI-based scrap detection powered by navio VISION reduces misclassification rates by 50%, improves overall quality control, and enables better process tuning based on detailed error classification.
:quality(80))