Machine Learning for Corrosion Prediction and Assessment of Reinforced Concrete Structures: Methods, Advances, and Challenges

During their service life, reinforced concrete structures are prone to reinforcement corrosion due to factors such as chloride salt attack and concrete carbonation, which lead to a decline in the load-bearing capacity and durability of the structural members. Traditional corrosion prediction models struggle to characterize the nonlinear behavior of corrosion evolution under the interaction of multiple factors, and conventional electrochemical testing also fails to fully capture the spatial distribution and long-term evolution patterns of corrosion; machine learning thus offers a new technical approach to address these challenges. Machine learning has already been applied in the fields of predicting rebar corrosion indices, assessing damage conditions, and evaluating the durability of corroded structural members. At the same time, various mechanism-integrated approaches—such as simulated data augmentation, few-shot learning, and physics-informed neural networks—have been developed. However, several prominent issues remain, including a lack of standardized datasets, limited cross-engineering generalization capabilities of models, a disconnect between data-driven approaches and the physical mechanisms of corrosion, and difficulties in practical engineering implementation. This paper outlines the complete modeling process for using machine learning to predict and assess the corrosion of reinforced concrete, identifies the appropriate applications for shallow and deep learning algorithms, summarizes the integration paradigms between machine learning and traditional durability techniques, reviews research progress in three major application areas, systematically analyzes key bottlenecks in this field, and proposes future directions, with the aim of providing a reference for intelligent research and engineering applications related to the durability of concrete structures.