Iterative building condition assessment and prediction using unified 3D point clouds and a BIM model: A case study

Visual building condition assessment is resource-intensive, particularly when inspections are repeated across large areas. This study seeks to reduce this burden by predicting the visual condition of surface-level architectural and structural elements and the latent risk in underlying elements using condition graphs. The proposed Scan-to-Condition prediction workflow builds on Scan-to-BIM and extends it to support ongoing inspection and prediction. This is demonstrated in a case study of an 818 m 2 mixed-use building in Melbourne that lacked prior BIM documentation. The workflow consisted of 4 components: (1) an optimised terrestrial laser scanning protocol for repeatable documentation and manageable data volumes, (2) a Scan-to-BIM modelling schema enriched with element-level confidence metrics and hidden-profile inference (3) a unified, multimodal condition documenting platform, and (4) an automated BIM-to-Graph Markup Language (GraphML) conversion method that generated graph representations for condition prediction using a custom Graph Attention Network (GAT) model. The workflow was designed as an iterative cycle in which baseline and follow-up point clouds, a semantically enriched BIM, and inspector reports were progressively integrated to form a longitudinal dataset. Future predictions are proposed to drive inspections and maintenance, progressively reducing manual human inspections over time. • Aged buildings without existing drawings or 3D models can still be represented digitally using a Scan-to-Condition workflow. • The Scan-to-Condition process in this study involves 3D laser scanning of the building to generate a point cloud model, which is then converted Building Information Modelling (BIM) enriched with material information and integrated into a condition prediction framework. • An optimised laser scanning protocol was developed and validated to ensure repeatable and consistent surveys. • BIM and 3D point cloud data with condition assessment data were effectively integrated to support iterative and ongoing condition assessments. • A BIM-to-GraphML workflow was created to enable Graph Neural Network (GNN)-based condition prediction from building element level.