Machine Learning-Based Monitoring Technology for Oxygen Concentration in Subsea Tunnel Concrete

The reinforced concrete structures of subsea tunnels are long subjected to the synergistic effects of external chloride attack and internal oxygen diffusion, which easily induce differential oxygen corrosion and threaten structural durability. To address the limitations of existing monitoring technologies, such as poor stability and short service life in complex marine environments, this study develops a novel embedded sensor for oxygen concentration monitoring in concrete. Regarding structural design, the proposed sensor incorporates a multi-electrode array, a three-layer oxygen-permeable membrane configuration (including a PTFE hydrophobic layer), and a composite gel electrolyte architecture, collectively enhancing its adaptability to high-alkalinity and high-salinity environments. In the experimental investigations, the potentiostatic polarization method is employed to establish the mapping relationship between oxygen content and current density. Through comparative testing, the electrode combination comprising a stainless steel working electrode, a titanium mesh reference electrode, and a stainless steel counter electrode was identified as the optimal configuration, achieving a correlation coefficient of 0.991. To validate system robustness under fluctuating service conditions, single-factor perturbation experiments were conducted across temperature (10~50 ℃), pH (7~14), and salinity (10.65~54.01 ms/cm) ranges, confirming the sensor’s reliability under multi-factor coupled environments. In terms of predictive modeling, an AdaBoost ensemble learning framework combined with SHAP (SHapley Additive exPlanations) interpretability analysis was introduced to construct a deep prediction model for oxygen content in complex dynamic environments. SHAP analysis results indicate that current density serves as the dominant predictor for oxygen content (with a SHAP value of approximately 0.8), while temperature constitutes the critical compensatory factor requiring interference correction. This research achieves continuous and precise monitoring of internal oxygen content within subsea tunnels, providing reliable technical support for the corrosion risk assessment and health management of offshore engineering structures.