A systematic review of machine learning advances in predicting pitting corrosion fatigue of marine and offshore structures

Pitting corrosion-fatigue remains a dominant degradation mechanism in offshore and marine steel structures. Conventional approaches for modelling pitting corrosion-fatigue often fail to capture the stochastic, localised progression of pit-induced fatigue under complex environmental and loading conditions. This study focuses on recent advancements in data-driven methodologies, particularly machine learning (ML) and hybrid models. ML frameworks, ranging from neural networks to ensemble learning, show considerable promise in modelling non-linear, high-dimensional relationships among corrosion, fatigue, and environmental parameters. Hybrid models, which integrate domain knowledge through physics-based features or mechanistic coupling, further enhance predictive accuracy and robustness while offering improved interpretability. This study highlights both opportunities and persistent challenges in the field, including the scarcity of standardised, high-fidelity datasets; difficulties in generalising models across different steel grades and exposure conditions; and the limited interpretability of certain ML algorithms. This study emphasises the development of intelligent, adaptable models to support improved structural health monitoring and asset management in offshore and marine environments.