Probabilistic Remaining Useful Life Estimation for Buried Pipeline Pitting Corrosion via Mechanics-Regularized Limit-State Learning

Buried steel pipelines are susceptible to external pitting corrosion whose spatially heterogeneous and time-dependent nature makes probabilistic remaining useful life (RUL) estimation both necessary and difficult. This study develops a three-stage framework comprising statistical pit depth characterization, a proxy corrosion-growth learner, and a mechanics-regularized limit-state RUL model anchored to the modified B31G burst criterion and an auxiliary conservative depth-screen rule, in which a corrosion-growth prior learned from sparse field measurements is embedded directly into the learning objective as a regularizer rather than merely used to construct training labels, and conformal calibration is applied at both the proxy and the limit-state stages to ensure honest empirical coverage, with the mechanics-regularized limit-state model consistently achieving coverage close to or above the nominal 90% target. Applied to the Velázquez excavation dataset and further examined on a 127-sample engineering database used as an auxiliary transferability check, the framework delivered stable point accuracy across all three wall thickness scenarios while maintaining empirical 90% coverage consistently above the nominal target. These results demonstrate that embedding mechanics-based growth constraints into the learning objective improves predictive consistency under sparse field data, while the resulting reliability outputs are best interpreted as scenario-based screening evidence for comparative integrity prioritization rather than as literal asset-life certification.