Tool wear in stamping processes has a significant impact on productivity and component quality. Optical monitoring offers the advantage of directly identifying wear and extracting process knowledge from the visual representation of wear phenomena. Supervised learning is usually associated with increased effort when creating a data set that only covers a limited spectrum of possible wear. The core objective of the present task is therefore the investigation of unsupervised learning approaches for the optical analysis of wear on blanking tools.

The following work packages are planned for this purpose:

• Research unsupervised learning approaches for image segmentation
• Selection, implementation and optimization of an approach for three existing data sets of different material combinations
• Interpretation and analysis of the results using techniques of Explainable AI
• Scientific documentation of the results