Image Recognition Using AI

Starting point How can defective parts be detected through image classification?

In body and vehicle manufacturing, a variety of workpieces are joined together using solder joints. Solder joints are defective if, for example, they contain cracks or pores. The suitability of solder joints can largely be assessed visually. Can the evaluation of images of solder joints be carried out sufficiently well and quickly enough using artificial intelligence? It should be noted that in the training images for image classification, defective solder joints are clearly underrepresented compared to high-quality solder joints.

What to do Solution: Classification of images by a neural network

The images of the solder joints are suitably cropped and centered. Appropriate data augmentation is used to enrich the imbalanced dataset. A neural network is trained on the training dataset, which classifies solder joints into different categories based on the images (“good”, “contains crack”, “contains pores”, “contains chip residues”, …). The artificial intelligence created in this way is integrated directly into the production process as an embedded system together with a camera solution. The overall system is designed so that expert feedback regarding classifications and categories can be incorporated, thus creating a gradually improving, self-learning system.

Your benefits Benefits: Savings in costs, time, and resources

• Time savings during quality checks
• Reduction of production costs
• Quality improvement in production by avoiding human errors and omissions in monotonous tasks

Further information Video material and further reading

• The Region Mainfranken GmbH reports on a computer vision project with our customer IFSYS, a manufacturer of feeding systems from Großbardorf, in their “Best Practices” article series, which highlights collaborations within their regional MaKoMA network: Mechanical engineering meets AI: a network with future prospects, January 2024 (in German). In this solution, neural networks have been integrated into a hazelnut sorting machine constructed by IFSYS, which, thanks to artificial intelligence, reliably sorts hazelnuts into different quality classes with high accuracy and speed.
• Image recognition with AI doesn’t have to be complicated or highly complex. In this video, we demonstrate how image classification can be achieved with minimal effort, low hardware requirements, and in a short amount of time – using screwdriver bits as an example.

References & Success Stories References