Machine-Learning Safety Screening of Lining Cracks Under Uncertain Tunnel Operating Conditions

Abstract

The structural integrity of transportation tunnels is paramount to ensuring the safety and efficiency of global infrastructure networks. Over their operational lifespan, tunnel linings are subjected to complex, dynamic, and often unpredictable environmental and operational conditions. These uncertain operating conditions accelerate the initiation and propagation of lining cracks, posing severe risks of structural failure. Traditional manual inspection methods and conventional non-destructive testing are labor-intensive, subject to human error, and often fail to accurately categorize crack severity under variable ambient conditions such as fluctuating illumination, groundwater seepage, and traffic-induced vibrations. This paper presents a comprehensive investigation into the application of machine-learning techniques for the automated safety screening of tunnel lining cracks specifically tailored to address uncertain operating conditions. A robust framework is conceptualized that relies on advanced feature extraction and classification algorithms capable of quantifying uncertainty in their predictions. By analyzing an extensive simulated and real-world dataset representing diverse tunnel degradation states under varying environmental stressors, the study demonstrates the efficacy of probabilistic machine-learning models in maintaining high diagnostic accuracy. The findings suggest that incorporating uncertainty quantification into safety screening pipelines significantly reduces false-negative rates in crack detection, thereby providing a more reliable foundation for predictive maintenance and infrastructure lifecycle management.