Machine learning-driven prediction of chloride resistance and service life estimation in blended cement concrete

초록

This study presents a methodology for durability-based design using machine learning (ML) models to predict chloride resistance and service life in blended concrete. Key ML models—Gaussian Process Regression (GPR), Series Neural Networks (SNN), and ensemble methods—were employed to estimate chloride migration and diffusion coefficients for concrete containing slag, fly ash, and silica fume. GPR and SNN models achieved the highest accuracy across datasets, with each model demonstrating optimal performance in specific chloride exposure conditions. The ML models’ predictions aligned conservatively with experimental data and fib Model Code 2010 values, reinforcing their reliability. Probabilistic simulations revealed that ML-predicted migration coefficients significantly influence service life estimates, particularly for slag and fly ash binders. The findings suggest that pretrained ML models can support early-stage durability assessments, supplementing traditional design methods when experimental data are limited.

키워드