Physics-Enhanced Machine Learning for Prediction of Chloride Diffusion in Concrete

Chloride-induced corrosion constitutes a considerable threat to the longevity of concrete infrastructure. This study introduces a physics-enhanced machine learning framework that incorporates chloride diffusion mechanisms into XGBoost via multi-level feature engineering. Utilising a dataset comprising 624 samples, we establish physics-based, material, and durability features derived from Fick’s law, the Arrhenius equation, and foundational principles of concrete science. The proposed model attains an R2 value of 0.868, surpassing the performance of the purely data-driven XGBoost by 5.9% and decreasing the RMSE by 14.3%. The physics-enhanced approach demonstrates superior generalization under extreme conditions, offering an interpretable framework for durability prediction.