Integrated BIM-IoT framework for real-time construction safety risk management

Construction safety remains a persistent global challenge, with accidents causing severe human and financial losses. Traditional risk management methods, relying on static plans or isolated monitoring, fail to deliver timely and context -aware responses. While Building Information Modeling (BIM) provides digital representation and the Internet of Things (IoT) enables real-time sensing, prior studies rarely achieve deep integration of these technologies with Artificial Intelligence (AI) for proactive safety management. This paper proposes an integrated AI-driven BIM-IoT framework comprising four layers: multi-source data acquisition, BIM-based semantic enrichment, AI-powered risk perception and predictive analytics using attention-based deep learning, and decision-feedback mechanisms enhanced by Bayesian inference. A dynamic risk index model is formulated to fuse sensor, environmental, and task data, while the anomaly detection module leverages attention - based neural networks to enhance interpretability. Experimental evaluations on a mid-rise construction project with 2.1 million sensor readings show that the framework outperforms baselines, improving prediction accuracy by 14.4% over BIM - only and 6.7% over IoT-only, while reducing latency to 1.4 seconds and halving false alarms. Ablation experiments confirm the contributions of semantic enrichment, Bayesian calibration, and interpretability modules, and robustness tests show resilience under sensor data loss. These findings advance BIM-IoT-AI research from interoperability toward functional coupling, offering both theoretical contributions and practical tools for intelligent, real-time construction safety management.