Towards Predictive Tunnel Safety: Integrating a Wireless Monitoring System with Machine Learning for Real-Time Risk Assessment

Summary

Abstract

The dynamic and uncertain nature of ground conditions poses a continuous challenge to safety and efficiency in tunnel construction. Traditional monitoring methods often involve time-lagged data acquisition, limiting the ability for proactive risk management. This paper presents a novel, integrated approach that leverages a advanced wireless sensor network (WSN) and machine learning (ML) to transition from reactive monitoring to predictive safety control. We have developed and deployed A robust WSN that provides continuous, real-time data on key geotechnical parameters directly from the rock mass and primary support. This system delivers critical information on deformation, stress, and environmental factors like water ingress directly to the surface, enabling immediate situational awareness. The core innovation lies in the application of bespoke ML algorithms to this rich, high-frequency datastream. These models are trained to identify complex patterns and precursors to instability that are often imperceptible through conventional analysis. The system successfully predicts evolving safety levels by forecasting potential adverse events, including excessive convergence, inadequate primary support performance, and sudden water inflows. By providing early warnings, this predictive capability allows engineers to implement timely and targeted countermeasures, such as adjusting support systems or deploying drainage, before critical conditions develop. This paradigm shift from observation to prediction significantly enhances operational safety, mitigates the risk of collapse, and ensures the structural integrity of the tunnel throughout its construction and service life. The presented framework demonstrates a significant step forward in data-driven, intelligent tunnelling.