Automated Generation of Disassembly Graphs Based on Detailed 2D Section Drawings

Abstract: Disassembly and circularity assessments are limited by the lack of structured, machine-readable data on material layers and connections, which are typically represented in expert-readable 2D section drawings. This paper explores how disassembly-relevant information can be automatically derived from such drawings represented as graphs using computer vision for segmentation, vision language models for material labeling, and language-based reasoning for semantic enrichment. The prototype extracts components and their connectivity and material information and generates graph representations capturing assembly hierarchies, enabling automated disassembly reasoning. This work marks a first step toward integrating heterogeneous data sources into unified graph models for circularity assessment.