Maintenance planning for aging reinforced concrete (RC) infrastructure is commonly based on deterministic service-life predictions derived from chloride ingress models. In these models, apparent chloride diffusion coefficients ( D app ) are treated as fixed inputs, even though they are inferred from experimentally variable data. This practice implicitly assumes stability in the timing of corrosion initiation and the need for intervention, an assumption that omits calibration-induced dispersion. Service-life prediction is reframed as a decision-risk problem by propagating uncertainty in calibrated diffusion parameters through corrosion initiation, maintenance, and life-cycle models. Machine-learning models are used to infer D app from published experimental datasets, and the associated predictive uncertainty is propagated through probabilistic corrosion initiation and maintenance models. Distributions of initiation time are mapped to maintenance triggers, enabling evaluation of variability in intervention timing, greenhouse gas emissions, and life-cycle costs. Results show that incorporating diffusion-parameter uncertainty increases expected maintenance demand by approximately 20–40% relative to deterministic predictions, whereas life-cycle environmental and economic impacts are up to 1.5–2 times the deterministic estimates across the 5 th –95 th percentile range. Decision sensitivity is most pronounced near common planning horizons, where marginal shifts in initiation distributions determine whether interventions are required. The findings demonstrate that deterministic service-life models may convey misleading stability in maintenance planning for RC structures. The proposed framework provides a decision-oriented methodology for integrating uncertainty in calibrated durability parameters into maintenance strategy evaluation, supporting more robust intervention planning for aging concrete infrastructure. This study quantifies how calibration-induced uncertainty in D app propagates through service-life and maintenance decision models. Treating calibrated inputs as fixed values underestimates intervention frequency and life-cycle impacts, particularly near common planning horizons.
Flah et al. (Sun,) studied this question.