Subway tunnels are critical elements of urban rail transit infrastructure. However, they are prone to external disturbances from adjacent construction activities and sustained operational loads, which can induce uneven settlement and local structural deformation. Hence, deformation monitoring is imperative for ensuring subway operational safety. In this paper, we address the limitations of conventional monitoring techniques, such as low frequency and vulnerability to instrument vibration, and the gap between data collection and structural analysis. Specifically, this paper proposes a framework, which integrates visual-inertial measurement and large language models, termed VILL, for comprehensive deformation and health monitoring of subway tunnels. In the VILL framework, a high-resolution optical imaging system integrated with an inertial sensor captures high-frequency data on monitoring targets and instrument pose, enabling high-precision deformation calculation. Simultaneously, a popular large language model is fine-tuned to analyze deformation data and automatically generate interpretable structural health assessment reports. Experimental results demonstrate that the VILL framework significantly suppresses environmental vibration noise while achieving high-frequency, high-accuracy monitoring coupled with automated analysis. This integrated system offers an end-to-end solution for real-time safety monitoring and early warning in tunnels and other critical transportation infrastructure.
Chen et al. (Fri,) studied this question.