Detection of human errors in construction structural design documents using automated extraction algorithm

The construction industry involves structural safety risks due to human errors arising from manual conversion and review across diverse software and document formats during design and construction phases. Although building information modeling (BIM) offers a collaborative framework, its adoption remains limited or partial on many sites. To address this limitation, this study proposes an artificial intelligence (AI)-based automated pipeline for detecting human errors in structural design documents without relying on BIM. The implemented AI includes Document Layout Network-You Only Look Once (DocLayNet-YOLO) for document layout analysis (DLA) and Surya-Optical Character Recognition (Surya-OCR) for text extraction from structural calculation and drawing documents. The application of AI in this study focuses on automated extraction, structuring, and cross-document consistency verification of reinforcement and dimensional information. In a case study, the proposed pipeline achieved information extraction accuracies of 95.56% for structural calculation documents and 90.47% for structural drawing documents. The system identified 253 human errors, including missing entries, main rebar mismatches, stirrup mismatches, dimension mismatches, and duplicate entries. These results demonstrate that the proposed AI-based pipeline can improve structural design review efficiency, reduce manual checking burdens, and support structural safety management in non-BIM construction environments.