AI for quality control in production

It is important to identify defective products quickly and efficiently and remove them before delivery in order to ensure that only high-quality products reach the customer. A prototype of such a system had already been implemented at the start of the cooperation with codecentric AG.

This involved taking pictures of the finished products, which were then evaluated by AI. This worked well to a large extent. However, there were significant variations in recognition performance, which could only partly be explained by the different product types. Furthermore, the metrics and warning indicators that sound an alarm whenever a deviation occurs were missing, so there was no counteraction against any algorithm malfunction at an early stage.

Working together in partnership, a variety of solutions were developed to meet the challenges. In a first step, advanced AI methods were used to make the system more robust in the face of various influencing factors such as product type or lighting. This immediately reduced the amount of work involved for manual checks. 

The second step was to operationalize the system for the factory floor. In the course of scaling up to five machines, the versions of the recognition software were initially managed centrally, which also enabled remote control and deployment of this software in the systems. A history recording function that is closely integrated with the system also allows in-depth analysis and evaluation. Integration with the company's existing monitoring software means that the responsible staff members are immediately alerted in the event of any system malfunction.