Image Recognition with the Help of Artificial Intelligence

How can defective parts be recognized through image classification?

In body and vehicle construction, a large number of workpieces are connected by means of solder joints. Solder joints are defective if they contain breaks or pores, for example. The suitability of soldered joints can largely be assessed optically. Can images of the soldered joints be assessed sufficiently well and sufficiently quickly using artificial intelligence? It should be noted that defective solder joints are clearly underrepresented in the training images compared to high-quality solder joints.

Solution: categorization of the images using a neural network

The images of the solder joints are suitably cut and centered. The imbalanced data set is enriched by means of suitable data augmentation. A neural network is trained on the training data set, which classifies solder joints into different categories based on the images (“good”, “contains breaks”, “contains pores”, “contains chip residues”, …). The artificial intelligence generated in this way is directly integrated into the production process in the form of an embedded system together with a camera solution. The overall system is constructed in such a way that expert feedback regarding classification and categories can be included. In this way, a gradually improving, self-learning system is created.

Benefits: automatic defect recognition leads to cost and time savings

  • Save time for manual quality checks
  • Save production costs
  • Increase in production quality by avoiding human errors

Video material and additional information

  • Region Mainfranken GmbH reports on a computer vision project with our customer IFSYS, a manufacturer of feeding systems from Großbardorf, as part of their “Best Practices” article series, which highlights collaborations within their regional MaKoMA network: Maschinenbau trifft KI: ein Netzwerk mit Zukunftsperspektive (in German) (external link), January 2024. In this solution, neural networks have found their way into a hazelnut sorting machine designed by IFSYS, which reliably sorts hazelnuts into different quality classes with high accuracy and speed thanks to artificial intelligence.
  • Classifying images with the help of AI does not automatically have to be complicated or highly complex. In the video (in German), we use screwdriver bits to show how image classification can be achieved with little effort, short time requirements, and minimal hardware.

Get in touch

Are you interested in one of our use cases and would like to discuss it with Tina Geisberger? Contact her to find out how we can assist you.

Bild Tina Geisberger

Tina Geisberger

Senior Account Manager - Tina's passion and expertise are use cases and she is eager to work with you to specify which of our use cases fit your situation. Through her years of professional experience, she knows how important it is to listen carefully to find out how our predictive analytics solutions can simplify your day-to-day work. She is extremely solution-oriented and welcomes any challenge - what does yours look like?


prognostica GmbH
Prymstr. 3
D-97070 Würzburg
P: +49 931 497 386 0

Your partner for Predictive Analytics and Data Science.

You can find further information, among other things concerning data security, in our imprint and privacy policy.

Follow us!

© 2023 prognostica GmbH