Entwicklung einer Bewertungsmethode zur Evaluierung verschiedener Aufgabeeinheiten für die Sortierung von Textilien

This thesis presents an AI-based evaluation method designed to provide an objective analysis of textile feeding units. The background lies in the growing importance of textile recycling and circularity, increasingly driven by political initiatives such as the European Green Deal and the EU Strategy for Sustainable and Circular Textiles. For textile sorting plants, it is essential that materials are evenly spread and sufficiently spaced on the conveyor belt. This arrangement is largely determined by the feeding units, which place the textiles onto the belt and thereby shape the downstream sorting process. To assess their performance, a method based on instance segmentation with FastSAM was developed. For every visible textile, a segmentation mask is generated, pre-processed and transformed into measurable indicators. Two program variants were implemented, one for whole garments and one for shredded textiles. The difference lies only in the preprocessing steps, such as refining mask precision or filtering out fully contained sub-masks. The calculation of the indicators is identical for both program versions. In total, the program determines six key indicator that together provide a comprehensive picture of the material distribution. These include object count, object size, belt occupancy, spatial distribution across the belt width, distances between objects, and the number of overlapping items. The approach was validated on approximately 200 images under realistic operating conditions. Results demonstrate robust performance but also reveal certain limitations, for example in cases of occluded surfaces or large contaminations. Overall, the method proves to be a practical tool for data-driven analysis of feeding units and enables reliable comparisons between different feeding systems.