Optimizing disassembly planning for aluminium-framed windows via automated adjacency data generation using collision-based geometric analysis

The construction industry faces growing pressure to reduce its environmental impacts through circular economy strategies. Disassembly planning enables systematic material recovery at end-of-life (EoL) yet remains underdeveloped for multi-material building products such as aluminium-framed windows. In CAD-integrated disassembly planning, adjacency matrices encode part-to-part relations that govern disassembly reasoning, but their accuracy depends heavily on geometric representation choices. This study proposes and validates a collision-based adjacency generation workflow for aluminium-framed window assemblies implemented in Rhinoceros-Grasshopper. We applied Boolean volumetric intersection to Bounding Box and Convex Hull geometric simplification under tolerance-compensated scaling to derive binary adjacency matrices. We evaluated seventeen window assemblies spanning fixed and operable configurations, multiple manufacturers, and two corner joint typologies. We validated digitally generated matrices cell-by-cell against manually constructed ground-truth matrices across part, component, and product levels. Results show that Convex Hull simplification combined with micro-scaling consistently achieves the closest alignment with ground truth, while Bounding Box simplification systematically overestimates adjacency through covering overlap. Non-scaled configurations underdetect tolerance-sensitive contacts regardless of the simplification method. The findings demonstrate that adjacency detection accuracy depends more on representation choices than on assembly size and that adjacency matrices encode fabrication logic as structured design fingerprints. The validated workflow provides a scalable, representation-aware basis for integrating geometric modelling with disassembly sequence planning, digital twin validation, and circular economy-oriented building design.