Integrating digital twin and large language models for advanced tower crane monitoring

Traditional monitoring approaches for tower crane operational safety primarily rely on manual inspections and univariate sensor threshold alarms, which exhibit significant limitations including delayed dynamic response and insufficient risk prediction capabilities, failing to meet real-time safety requirements in complex construction scenarios. To address these challenges, this study proposes an innovative intelligent monitoring system that integrates digital twin technology with multimodal large language models (MLLMs). The system first constructs a 3D digital twin of the crane using IoT-enabled digital twin technology, establishing multidimensional dynamic mapping between physical entities and virtual models to create a comprehensive digital representation encompassing mechanical structures, motion trajectories, and environmental parameters. Building upon this foundation, a multimodal MLLM-based analytical framework is designed to intelligently process surveillance video streams and identify potential safety hazards. The system employs a microservices architecture to develop a web-based visualization platform that integrates real-time situational awareness, abnormal behavior detection, operational status monitoring, and early warning functionalities. Experimental results demonstrate the system's capability to monitor crane operations in real time while effectively identifying potential risks and anomalies. The research contributes novel methodologies for digital twin construction, multimodal cognitive model architectures, and virtual-physical fusion warning mechanisms, providing both theoretical foundations and practical solutions for advancing safety management systems in construction sites.