Abstract: Water conveyance tunnels are vital infrastructure in major water diversion projects, playing a key role in ensuring the project's water supply capacity. To overcome the limitations of traditional single-layer linings—such as insufficient anti-seepage and crack resistance—an innovative prestressed double-layer lining structure has been developed for use in high-water-pressure environments. This structure employs a load-sharing mechanism: the outer shield segment liner withstands external hydrostatic and geostatic pressures, while the inner prestressed reinforced concrete liner controls crack development. However, long-term operation has revealed significant challenges, including material aging, performance degradation of seepage control and drainage systems, and complex interactions with environmental loads. These issues underscore the urgent need to establish a comprehensive safety assurance technology system. 　　The Yellow River Crossing Tunnel design has proven effective in withstanding high water pressure; however, several operational challenges have surfaced, including clogged drainage cushions, degradation of polyurea waterproofing materials, and localized defects such as cracks and voids. Key challenges include: (1) External threats—such as riverbed scouring, sediment deposition, and landslide-prone slopes—combined with internal factors like concrete aging and prestress loss, give rise to safety hazards that are both time-dependent and spatially uncertain. (2) Maintenance constraints: restricted shutdown periods and dependence on conventional repair techniques (e.g., epoxy grouting, steel plate bonding) often address only surface-level defects, while failing to remediate deeper issues such as longitudinal cracking and persistent drainage obstructions.　　Early risk assessment methods primarily employed qualitative techniques, such as expert surveys and fault tree analysis. Recent developments have introduced quantitative approaches, including the Analytic Hierarchy Process (AHP), Bayesian Networks (BN), and hybrid models like AHP–cloud models. For double-layer linings, critical risk factors include cracking, seepage, material degradation, and variable hydrostatic pressure. Dynamic BN models have been employed to capture the temporal evolution of risks, exemplified by observed correlations between seepage volume and temperature in the Yellow River Crossing Tunnel. Meanwhile, knowledge graph technologies are emerging as tools for integrating multi-source data—such as monitoring records and maintenance logs—to enable dynamic risk mapping, although their application in tunnel safety remains at an early stage. 　　Research on the mechanical behavior of double-layer linings has primarily concentrated on load distribution mechanisms and the effects of structural defects. Finite element analyses indicate that the stiffness of the drainage cushion plays a pivotal role in load transfer between the outer and inner linings: lower stiffness alleviates stress on the inner lining but may intensify stress concentrations at segment joints. Experimental investigations have delineated four typical failure stages under internal water pressure: linear elastic response, cracking of the inner lining, crack stabilization, and damage at segment joints. Notable findings include the critical influence of drainage cushion blockage in elevating seepage pressure and the heightened vulnerability of the tunnel crown due to insufficient concrete thickness. Nonetheless, existing studies predominantly adopt static, cross-sectional perspectives, with limited attention to longitudinal deformation and coupled multi-physical processes, such as seepage–stress–temperature interactions.　　Monitoring systems for double-layer linings primarily measure seepage volume, piezometric pressure, joint displacements, and structural stress/strain. While traditional models developed for single-layer tunnels—such as statistical regression and hybrid approaches incorporating structural calculations and thermal effects—have been adapted, they generally lack robust spatiotemporal analytical capabilities. For defect detection, non-destructive testing methods including ground-penetrating radar, laser scanning, and deep learning-based image analysis (e.g., CNN and Faster R-CNN) have demonstrated potential in identifying cracks and voids. However, challenges remain in mitigating signal noise and achieving real-time three-dimensional visualization. Safety evaluations are still predominantly qualitative, guided by industry standards (e.g., SL/T 790-2020), with limited application of data-driven, intelligent evaluation methods. 　　Sensor layout optimization for double-layer linings remains largely dependent on engineering experience, with limited research addressing robustness against environmental noise and sensor failures. Although reliability-based methods—such as Monte Carlo simulations and response surface techniques—have been employed to estimate failure probabilities under uncertain loading conditions, comprehensive system-level reliability models that incorporate spatial correlations among defects are still lacking. Condition-based maintenance (CBM) strategies, which aim to balance structural safety with continuous water supply, remain underdeveloped; maintenance decisions are predominantly guided by expert judgment rather than algorithmic assessment. Emergency response systems, despite their critical importance, are generally case-specific and have yet to incorporate knowledge-driven, intelligent scenario-generation capabilities.　　Future research should address several critical issues: (1) Multi-field coupling mechanisms: Current understanding of long-term structural performance under the combined effects of seepage, mechanical stress, temperature variation, and chemical degradation remains limited. (2) Data–mechanism fusion: Existing models tend to rely either on empirical formulations or data-driven algorithms, with insufficient integration of mechanistic insight and data analytics to support accurate safety evaluation. (3) Reliability-based maintenance: There is a lack of multi-objective optimization models for condition-based maintenance (CBM) that simultaneously consider safety assurance, cost-effectiveness, and water supply reliability. (4) Intelligent materials: Development is needed for self-healing materials—such as microcapsule-based crack repair systems and shape memory alloy prestressing technologies—that can autonomously address structural defects without requiring service interruptions. 　　Future research should focus on evaluating long-term structural behavior under dynamic loading conditions through advanced numerical models—such as coupled seepage–stress finite element methods (FEM)—supported by long-term monitoring data. Hybrid modeling approaches that integrate mechanistic analysis (e.g., FEM) with machine learning techniques (e.g., deep neural networks) should be developed for real-time safety assessment. Predictive models based on Markov chains, combined with multi-objective genetic algorithms, are needed to forecast degradation trajectories and optimize maintenance scheduling. Additionally, the development of self-sensing and self-repairing materials, including self-healing concrete and smart prestressing systems, will be essential to enhance the resilience of tunnel structures.　　Double-layer lined water conveyance tunnels have significantly improved the safety of high-pressure water transfer systems. However, their long-term performance remains challenged by aging infrastructure, complex environmental interactions, and restricted maintenance access. Addressing these challenges requires the development of integrated, data-driven frameworks encompassing risk identification, performance prediction, intelligent diagnostics, and proactive maintenance. By leveraging emerging technologies such as knowledge graphs, digital twins, and intelligent materials, future advancements will strengthen the safety and resilience of national water network infrastructure, ensuring a sustainable and reliable water supply for large-scale strategic projects.