“The high quality of our products is a matter of course at AT&S. Photos of the circuit boards are taken automatically during the manufacturing process and these are then run through image analysis software. Sometimes it happens that circuit boards are mistakenly recognized as ‘faulty’. Unfortunately for no reason we can understand. That cost us additional time and resources,” said Ulrike Klein, Head of the Data & Analytics department at AT&S.
“It was our aim to precisely identify the faulty circuit boards and to make the results comprehensible. We are pleased that we were not only able to implement the project successfully, but that our results also agree with the statements of the AT&S technicians”, says Dr. Andreas Trügler, Head of the DDAI Module at the Know-Center. “First our algorithm had to understand which circuit boards were faulty and why. To do this, the team trained a neural network and fed it with image data from correct and faulty circuit boards. Using methods from the ‘Explainable AI’ research field, we were also able to provide an explanation of why and where a circuit board was identified as defective.“
Intelligent production using AI
“AI enables quality assurance at the highest level and saves companies costs and resources. Automated image recognition and analysis, which is finding its way into many industries, still show quality gaps. Another obstacle to firmly anchoring AI in companies is trust in these technologies. We are very pleased that we succeeded in overcoming both barriers in this project“, Stefanie Lindstaedt, Managing Director of the Know-Center explained.
