Uncertainty-aware prediction of chloride resistance in recycled aggregate concrete exposed to marine conditions

Recycled aggregate concrete (RAC) supports circular‐economy and carbon-reduction goals, yet its chloride resistance is often impaired by adhered mortar and microcracking in recycled aggregates, creating uncertainty for durability prediction—particularly in chloride-rich marine environments. This study adopts the electric flux Q measured by the ASTM C1202 rapid chloride permeability test (RCPT) as the durability indicator and develops a Bayesian–MCMC prediction framework that yields full posterior distributions for model parameters and predictive intervals for Q . By combining likelihoods of measured fluxes with informed priors, the framework quantifies contributions from material variability, measurement noise, and model bias within a single, interpretable model. Validation against a comprehensive laboratory dataset shows strong calibration: 86% of predictions lie within ±30% of measured Q ; 95%/50% credible bands achieve 96.3%/50.7% coverage, indicating uncertainty bounds that reflect—not understate—experimental scatter. Diagnostics across mixture variables confirm bias-controlled, stable predictions within the calibrated domain. Sensitivity analyses identify the water-to-binder ratio ( w / b ) as the first-order driver (median global standardized importance ≈ 50%), followed by Ground Granulated Blast-furnace Slag (GGBS), fly ash, recycled aggregate replacement, and age (each ~10–15%); residualized absorption contributes <5% once RA content is accounted for. Two-variable partial-dependence surfaces reveal design-relevant interactions: SCM benefits are greater at lower w / b , curing exhibits diminishing returns at high w / b , and GGBS partially offsets RA’ adverse effect, enabling compensatory mix strategies. The resulting tool provides predictive means and credible intervals that can be mapped directly to RCPT-based acceptance probabilities (e.g., the probability that Q meets project limits), supporting performance-based mixture optimization and uncertainty-aware service-life assessment for RAC in marine and coastal infrastructure. • Bayesian–MCMC predicts RCPT flux with calibrated uncertainty. • Bridges RAC sustainability and marine durability. • Mechanistic drivers: w / b dominant; SCM synergy; RA mitigable. • Supports performance-based mix design and acceptance decisions.