Mining of Reusable Component Libraries through Unsupervised Multi-Modal Clustering
Master's Thesis — Mustafa Sercan Amaç (2026)
Student: Mustafa Sercan Amaç
Type: Master’s Thesis in Informatics
Submitted: June 8, 2026
Supervision / advising team: Panagiotis Petropoulakis, Georgios Pavlidis, André Borrmann, Alois C. Knoll
This thesis develops an unsupervised multimodal pipeline for mining reusable component libraries from large BIM/IFC repositories. It combines handcrafted geometric features, language-model-generated descriptions and embeddings, and visual foundation-model embeddings, followed by dimensionality reduction, clustering, and multimodal retrieval.
The work investigates how latent component structure can be recovered beyond noisy or overly broad IFC type labels, enabling more effective search, reuse, and curation of building components.