We designed a hardware demonstrator for exhibition of AI image analysis. It con-sists of a microscope, a portable GPU-PC, and a screen. Using this setup, it is possible to explore the usefulness and limitations of the AI algorithm for image analysis.

In this project, three priorities were set which are briefly described here: A flexible AI algorithm for image analysis was optimized for the use case of tool wear detection. This algorithm was improved with regards to image size, network size, and image augmentation. These optimizations were then utilized in a new use case: AI-based detection of melt pool area on microscopic cross-section images of seams in laser beam welding, as shown in Figure 1. Furthermore, the indirect estimation of the hard-to-determine keyhole depth in laser beam welding using the measured surface roughness of the seams was investigated. In the first approach, which can be seen in Figure 2, no useful correlation for indirect quality assessment could be found.