Improved small-sample estimation method and stochastic analysis of concrete durability parameters in marine environment

To address the small-sample estimation problem of concrete chloride ingress parameters—apparent diffusion coefficient ( D app ) and surface chloride concentration ( C s )—in marine environments, an Adaptive Robust Hybrid Estimator (ARHE) is proposed. This method integrates Quantile Matching (QM) and Weighted Maximum Likelihood Estimation (WMLE) through a two-stage strategy to enhance robustness: the QM stage utilizes the median and interquartile range (IQR) to resist outliers; the WMLE stage introduces the Huber weight function for dynamic parameter optimization. Based on concrete exposure tests in a simulated seawater environment (60d/120d/240d, n = 30 per age group), ARHE was compared with traditional Maximum Likelihood Estimation (MLE) and Unbiased Estimation (UE) methods. Results indicate that as exposure time increased from 60d to 240d, the mean C s rose from 0.43 % to 0.56 %, while the mean D app decreased from 7.85 × 10 −12 to 3.45 × 10 −12 m 2 /s, consistent with hydration densification mechanisms. ARHE significantly reduced variance estimation error: the variance of C s was reduced by 44.3 %–96.1 % compared to MLE and UE, and the variance of D app was reduced by 71.9 %–94.5 %. ARHE's outlier resistance improved, with parameter fluctuation amplitudes only 5.3 %–79.8 % of those from traditional methods. C s follows a normal distribution, while D app follows normal or log-normal distributions. The coefficient of variation (CV) of C s decreased from 18.2 % at 60d to 7.1 % at 240d, reflecting reduced stochasticity. ARHE provides a highly robust approach for estimating concrete durability parameters under small-sample conditions. • ARHE integrates quantile matching and weighted MLE to suppress outliers and small-sample bias. • Variance errors for C s / D app reduced by 44–96 % vs. MLE/UE ( n = 30). • C s follows normal distribution; randomness weakens (CV 18.2 %→7.1 %). • ARHE improves input reliability for chloride-ingress models and service life predictions.