Artificial Intelligence-Driven Quality Assurance for Automated Detection of Structural Engineering Errors and Constructability Risks in Digital Infrastructure Models

Abstract

Digital infrastructure delivery increasingly depends on building information modelling (BIM), Industry Foundation Classes (IFC), structural analysis models and digital-twin workflows, yet model quality assurance remains heavily dependent on manual review and conventional clash detection. This report develops and evaluates a hybrid artificial intelligence-driven quality assurance (AI-QA) framework that combines semantic and geometric IFC validation, deterministic structural rules, machine-learning classification, anomaly detection and constructability-risk prioritization. The supplied evaluation dataset contains 1,000 labelled model-quality conditions distributed across geometry, topology, structural properties, loads and boundary conditions, semantics, model completeness, interdisciplinary clashes, constructability/accessibility and construction-sequence risk classes. Comparative results show that the hybrid framework achieved 96.4% accuracy, 95.8% precision, 96.1% recall, 95.9% F1-score and 0.98 ROC-AUC, exceeding conventional clash detection, rule-based BIM checking and standalone machine-learning baselines. Interdisciplinary clashes recorded the highest class-level F1-score (98.3%), while construction-sequence risks remained the most difficult class (88.7%). Model-review time decreased from 16.5 h for manual review to 2.7 h, an 83.6% reduction. Detection performance also increased with BIM information completeness, reaching 96.4% at approximately 95–100% completeness. The findings support a human-supervised, explainable AI architecture in which automated screening and risk prioritization augment, rather than replace, structural engineering judgment. The report concludes with recommendations for IFC 4.3/ISO 16739-1:2024 interoperability, machinereadable information requirements, project-level validation and continuous learning from validated QA/QC records.